Development of an AI-derived, non-invasive, label-free 3D-printed microfluidic SERS biosensor platform utilizing Cu@Ag/carbon nanofibers for the detection of salivary biomarkers in mass screening of oral cancer

Abstract

Developing a non-invasive and reliable tool for the highly sensitive detection of oral cancer is essential for its mass screening and early diagnosis, and improving treatment efficacy. Herein, we utilized a label-free surface enhanced Raman spectroscopy (SERS)-based biosensor composed of Cu@Ag core–shell nanoparticle anchored carbon nanofibers (Cu@Ag/CNFs) for highly sensitive salivary biomarker detection in oral cancer mass screening. This SERS substrate provided a Raman signal enhancement of up to 10⁷ and a detection limit as low as 10⁻¹² M for rhodamine 6G molecules. Finite-difference time-domain (FDTD) simulation studies on Cu@Ag/CNFs indicated an E-field intensity enhancement factor (|E|²/|E₀|²) of 250 at the plasmonic hotspot induced between two adjacent Cu@Ag nanoparticles. The interaction of this strong E-field along with the chemical enhancement effects was responsible for such huge enhancement in the Raman signals. To realize the real capability of the developed biosensor in practical scenarios, it was further utilized for the detection of oral cancer biomarkers such as nitrate, nitrite, thiocyanate, proteins, and amino acids with a micro-molar concentration in saliva samples. The integration of SERS substrates with a 3D-printed 12-channel microfluidic platform significantly enhanced the reproducibility and statistical robustness of the analytical process. Moreover, AI-driven techniques were employed to improve the diagnostic accuracy in differentiating the salivary profiles of oral cancer patients (n₁ = 56) from those of healthy controls (n₂ = 60). Principal component analysis (PCA) was utilized for dimensionality reduction, followed by classification using a random forest (RF) algorithm, yielding a robust classification accuracy of 87.5%, with a specificity of 92% and sensitivity of 88%. These experimental and theoretical findings emphasize the real-world functionality of the present non-invasive diagnostic tool in paving the way for more accurate and early-stage detection of oral cancer in clinical settings.

---

Introduction

Oral cancers are classified as the eighth most prevalent cause of cancer-related mortality globally. Oral squamous cell carcinoma (OSCC) accounts for over 90% of all oral cancer cases, and is increasingly recognized as a significant public health issue worldwide, given its high incidence and low survival rates.¹ Significant risk factors associated with the development of oral cancer include the consumption of alcohol, the use of tobacco products, and infection with human papillomavirus (HPV). Early detection is crucial for improving treatment outcomes, ensuring a better prognosis, and augmenting the quality of life for patients. Biopsy is recognized as the gold standard for diagnosing oral lesions. However, this procedure is invasive, technically intricate, and relatively expensive, often resulting in patient discomfort. Given these limitations, it is crucial to explore and evaluate alternative diagnostic modalities that are non-invasive, cost-effective, reliable, and accurate.²

Saliva stands out as a ground-breaking option in the realm of liquid biopsies due to its complexity compared to other biological fluids like urine, blood, sweat, and tears. Furthermore, saliva also reflects the level of biomarkers in real time, similar to plasma, and offers many benefits like non-invasive and painless nature, reduced costs, and ease of storage, which collectively position it as a promising candidate for diagnostic applications in clinical practices.^3,4 Saliva also represents a promising diagnostic tool for oral cancer due to the close correlation between salivary composition and serum metabolites. Many reports have shown that tumours can release biochemical compounds into saliva during tumour lavage, resulting in elevated concentrations of biomolecules that reflect cellular changes associated with neoplastic processes, which suggests that analysing salivary biomarkers may enhance the detection and diagnosis of oral cancer malignancies.^5,6 Key salivary biomarkers for the diagnosis of oral cancer include nitrate, nitrite, and thiocyanate. Recent studies have demonstrated that salivary concentrations of nitrate and nitrite are significantly elevated in patients with oral carcinoma compared to the control groups. This can be attributed to the presence of nitrosamines in the saliva of smokers and habitual tobacco chewers. These nitrosamines likely originate from tobacco or are generated endogenously through the nitrosation of secondary amines and tobacco alkaloids. The nitrosation process, which converts organic compounds into potent teratogens and carcinogens known as nitroso derivatives, is initiated by elevated levels of thiocyanate (SCN) ions in the saliva of smokers, thereby increasing their susceptibility to cancer. Salivary concentrations of thiocyanate (SCN) in healthy non-smokers typically range from 0.5 to 2 mM, with an average of 1 mM, though levels can increase to as much as 6 mM in heavy smokers. In the control group, salivary nitrite levels were observed to range from 30 to 100 μg L⁻¹. Conversely, oral cancer patients demonstrated significantly elevated nitrite levels, exceeding 200 μg L⁻¹.

A significant challenge in utilizing saliva for cancer diagnostics is its restricted availability in patients, particularly in individuals with lung cancer who may suffer from xerostomia (dry mouth syndrome), which is characterized by diminished salivary secretion.⁷ Conventional diagnostic methodologies often necessitate larger or multiple saliva samples, which can be impractical in specific contexts. To overcome this limitation, researchers have developed microfluidic systems capable of efficiently partitioning small saliva samples into several aliquots. These chips are engineered for minimal volume handling, thereby optimizing the utilization of available saliva for a range of analyses and enhancing the overall efficacy of salivary diagnostics.

The enzyme-linked immunosorbent assay (ELISA), high performance liquid chromatography (HPLC), microsatellite analysis, liquid chromatography–mass spectrometry (LC–MS), etc. are frequently used for salivary biomarker detection.^8,9 However, despite their potential to detect bio molecular changes, current diagnostic methods often necessitate expensive, time-consuming equipment and complex sample preparation procedures. Moreover, these techniques frequently lack the required specificity and sensitivity to meet stringent clinical standards for early tumour diagnosis. This underscores the need for a rapid, cost-effective, and highly sensitive diagnostic approach. In this scenario, surface enhanced Raman spectroscopy (SERS) emerges as a promising alternative, offering detailed fingerprint data on biomolecules that could facilitate the identification of critical changes relevant to medical diagnosis.^10–12 Plasmonic nanoparticles, such as silver, gold and copper have been widely used as SERS platforms due to the generation of hotspots caused by colloidal aggregation.¹³ However, the poor dispersity of nanoparticles in complex samples limits their wider applicability causing poor reproducibility. Bimetallic nanostructures leverage the combined physical and chemical properties of their constituent metals, presenting a highly effective strategy for the development of high-performance SERS substrates. This approach overcomes several limitations associated with single metal nanoparticles, such as elevated costs, insufficient SERS signal strength, and discrepancies in optical and electrical properties. Consequently, bimetallic systems enhance the overall efficacy of SERS applications.^14,15 The cost-effectiveness and remarkable versatility of core–shell structured nanoparticles, comprising an inner core and an outer shell made from diverse materials, have attracted significant attention in recent research. Silver exhibits a higher surface-enhanced Raman scattering (SERS) enhancement compared to other plasmonic metals, while metallic copper is characterized by its stability and biocompatibility. Consequently, Cu@Ag core–shell nanostructures represent a promising candidate for sensing applications in SERS.^16–18

To mitigate the challenges posed by nanoparticle aggregation, a variety of substrates have been examined as solid support materials for plasmonic nanoparticles, with the goal of enhancing hot-spot density and lowering the detection limit of SERS. Among these SERS substrates, electrospun carbon nanofibers are particularly advantageous due to their high surface area, which facilitates the accommodation of a high density array of plasmonic nano clusters, thus positioning them as a promising solid support for SERS applications.^19,20

This investigation demonstrates a cost-effective, label-free SERS based salivary biosensor integrated into a microfluidic platform, designed for the detection of oral cancer biomarkers employing AI algorithms to differentiate between oral cancer patients and healthy controls. To the best of our knowledge, this is the first report on the development of copper@silver core–shell nanoparticles supported on carbon nanofibers (Cu@Ag/CNFs) for use as SERS substrates for salivary biomarker sensors integrated into a microfluidic platform. The electrospinning technique was employed to fabricate carbon nanofibers through a two-step thermal treatment process, which included stabilization and carbonization and the bimetallic core@shell Cu@Ag/CNFs were fabricated through a combination of chemical reduction followed by transmetallation reaction. Furthermore, we examined the utilization of label-free SERS for the biochemical analysis of saliva samples obtained from patients diagnosed with oral cancer, as well as from healthy controls. SERS spectra from both cohorts were systematically analyzed and differentiated through a multivariate statistical approach, including random forest combined with principal component analysis (PCA-RF), which effectively distinguished between oral cancer groups and healthy controls. The findings underscore the potential of the developed plasmonic SERS material for facilitating a simple, cost-effective, non-invasive, and label-free diagnostic tool for the early detection and screening of oral cancer.

Experimental section

Materials and reagents

Polyacrylonitrile (PAN, average molecular weight = 150 000, Sigma Aldrich), poly vinyl alcohol (PVA, molecular weight = 89 000–98 000, Sigma Aldrich), dimethyl formamide (DMF, 99%, Merck), sulphuric acid (H₂SO₄, 98.08%, Merck), nitric acid (HNO₃, 70%, Merck), silver nitrate (AgNO₃, 99.5%, Merck), copper(II) sulphate pentahydrate (CuSO₄·5H₂O, 99%, Merck), sodium borohydride (NaBH₄, 97%, Merck), sodium hypophosphite monohydrate (NaH₂PO₂·H₂O, 99%, LOBA Chemie), and hydrazine hydrate (N₂H₄, 99 to 100%, Fisher Scientific) were used for the synthesis of Cu@Ag/CNFs. All reagents were of analytical grade and directly used without any further purification.

Preparation of Cu@Ag/CNFs

The carbon nanofibres were fabricated using the electrospinning technique as explained in our previous works.^21,22 Briefly, 8 wt% of polyacrylonitrile (PAN) was dissolved in DMF at 60 °C under vigorous stirring for 2 h. Using a variable high voltage power supply, a high voltage of 18–20 kV was applied during the electrospinning process with a flow rate of 0.5 mL h⁻¹ and a needle to collector distance of 21 cm. A thick fibrous mat was created by depositing the electrospun PAN fibres onto a clean copper sheet. The as-spun PAN nanofibres were transformed into carbon nanofibers through a two-step thermal process that involved stabilisation at 300 °C and carbonization at 800 °C under air and N₂ atmospheres, respectively, with a ramping rate of 5 °C min⁻¹. The obtained carbon nanofibers were then functionalized using a solution mixture of H₂SO₄ (3 M) and HNO₃ (3 M) (3 : 1 v/v) in acidic reflux conditions for 8 h at 120 °C. Following the acid functionalization, the resulting carbon nanofibers were separated, thoroughly washed in deionized water until neutral pH was obtained, filtered and dried overnight at 120 °C in a vacuum oven.

The synthesis of the bimetallic Cu@Ag core–shell nanoparticle-anchored carbon nanofibres was carried out using a two-step process consisting of a chemical reduction method for the synthesis of Cu core nanoparticles, followed by transmetallation reaction for the formation of a Ag shell over the Cu core nanoparticles. The copper anchored carbon nanofibers were prepared using the chemical reduction method with CuSO₄·5H₂O as the copper precursor and NaH₂PO₂·H₂O was used as the reducing agent for the reduction of Cu²⁺ ions to Cu⁰ nanoparticles.²³ In brief, 60 mg of f-CNFs were dissolved in 20 mL of ethylene glycol and sonicated for 30 min. The above solution was mixed with 0.5 g CuSO₄ and 0.3 g PVP to form solution A. Furthermore, 0.63 g NaH₂PO₂·H₂O was dissolved completely in 10 mL of distilled water to form solution B. Finally, solution B was added to solution A under constant stirring conditions. The reaction was allowed to proceed at 120 °C for 1.5 h with the molar ratio of N₂H₄ to CuSO₄ maintained at 3. The heater was turned off after the colour of the solution was changed from blue to reddish brown indicating the formation of copper nanoparticles. The mixture was then allowed to cool naturally to room temperature to extract the Cu nanoparticle-anchored CNFs. Furthermore, transmetallation reaction was carried out using 30 mM AgNO₃ as the precursor to create a silver shell on the surface of the core Cu nanoparticles anchored on the CNFs. The reduction process of silver nitrate was completed by adding ice-cold solution of 10 mM sodium borohydride drop-wise to the above solution after the reaction mixture was stirred for 1 h at room temperature. The resulting slurry was then filtered, thoroughly washed with distilled water, and dried overnight in a hot air oven at 80 °C to obtain the core–shell Cu@Ag/CNFs (1 : 3 ratio). A similar synthetic procedure was used for the synthesis of Cu@Ag core@shell nanoparticles with various bimetallic Cu : Ag ratios (1 : 1, 1 : 5) anchored over carbon nanofibers.

Fabrication of the SERS substrate

The glass slides were pre-cleaned several times with ethanol and ultrapure water and dried at 40 °C for 30 min. The SERS substrate was fabricated by dispersing 5 mg of the Cu@Ag/CNFs in ethanol through sonication and 20 μL of the above nano-composite dispersion was drop cast onto the surface of the glass slides followed by drying at 40 °C for 30 min. Rhodamine 6G was used as the standard probe molecule to evaluate the SERS efficiency of the Cu@Ag/CNF-based SERS substrates. The R6G solutions with concentrations ranging from 1 × 10⁻⁶ to 1 × 10⁻¹² M in DI water were prepared using the stock R6G solution (0.1 M). The SERS analysis of different concentrations of R6G was carried out by drop-casting 20 μL solution of R6G with specific concentrations ranging from 1 × 10⁻⁶ to 1 × 10⁻¹² M onto the developed Cu@Ag/CNF-based SERS substrate.

Saliva sample preparation

Saliva samples were collected from both healthy volunteers and oral cancer patients 1 h after food intake and after rinsing the mouth with pure water. Saliva was directly collected into a disposable cup using a non-stimulated collection method. In order to remove food debris and oral mucous epithelial cells, the saliva samples were transferred into 2 mL centrifuge tubes and centrifuged at 12 000 rpm for 15 min at 4 °C. The supernatant was collected after centrifuging to obtain pure saliva, which was then stored at 4 °C until analysis. Prior to SERS analysis, 0.5 mL of the saliva samples were mixed with 0.5 mL of deionized water and 10 μL of the as prepared saliva sample was drop cast onto the surface of the SERS substrate and subjected to atmospheric drying overnight.

Fabrication of a microfluidic chip for SERS analysis

The microfluidic chip was designed using the SOLIDWORKS 3D CAD design software. 3D printing was used for the fabrication of the microfluidic chip using the EP-A450 3D printer stereo lithography. It involves directing a UV laser (355 nm) onto a vat of photopolymer resin (polycarbonate). The pre-programmed design was drawn onto the photopolymer vat's surface using the UV laser. Since photopolymers are sensitive to ultraviolet light, the resin is photochemically solidified to form a single layer of the desired three-dimensional object. This procedure was carried out again for every design layer until the 3D object was finished. The developed microfluidic chip consists of a main well and 12 sub wells connected through channels.

AI approaches

Pre-processing of the raw spectra was necessary to eliminate experimental artefacts before statistical analysis. The raw SERS spectra were subjected to baseline correction, smoothing, and the elimination of the fluorescence background to clearly identify the characteristic Raman signals. The principal component analysis with random forest (PCA-RF) approach is an effective way to analyse biological Raman spectroscopic datasets. Principal components, or PCs, are a statistical method for transforming data from a higher dimensional space into a lower dimensional space. Twenty dimensions were used in the analysis. The random forest algorithm was used for classification following PCA, taking advantage of its robustness against over fitting and capacity to manage intricate feature interactions. The approach strikes a balance between computational efficiency and predictive performance by combining PCA and random forest, making it easier to distinguish oral cancer from healthy controls with accuracy and reliability.

Characterization techniques

The surface morphology of the Cu@Ag/CNFs was examined using scanning electron microscopy (SEM, Carl Zeiss) image analysis. X-ray diffraction (XRD) measurements were acquired using a PANalytical X-pert diffractometer (Malvern) with Cu Kα radiation (λ = 1.54 Å) as the source. Cu@Ag/CNFs were morphologically characterised by means of transmission electron microscopy (JEM-2010, 200 kV, JEOL, Japan) combined with EDAX. The microfluidic chip was fabricated using the EP-A450 3D printer stereo lithography, and SERS spectra were acquired using a confocal Raman microscope (WITec alpha300 RA, Ulm, Germany) equipped with a 532 nm laser.

Results and discussion

The bimetallic Cu@Ag core–shell nanoparticles were anchored on electrospun carbon nanofibers using chemical reduction followed by transmetallation reaction (Scheme 1). Since CNFs are typically smooth and chemically inert, activating their surfaces with acid refluxing is necessary before functionalizing them with Cu@Ag nanoparticles. This functionalization aids in the uniform anchoring of the core@shell nanoparticles to the CNF surface. Chemical reduction has emerged as an enthusiastic technique for the synthesis of size-controlled copper nanoparticles. In the first step, the chemical reduction method was used to develop a stable aqueous dispersion of core-copper nanoparticle anchored CNFs. Following the addition of the reducing agent, a change in the sol colour was observed, with the light-blue aqueous solution turning into wine red, indicating the formation of copper nanoparticles. The reduction reaction mechanism of Cu²⁺ by H₂PO₂⁻¹ could be expressed as:

Cu²⁺ + H₂PO₂⁻ + H₂O → 2Cu↓ +H₂PO₃⁻ + 2H⁺ (1)

Scheme 1 Synthesis process adopted for the fabrication of bimetallic Cu@Ag core@shell-anchored carbon nanofibers (Cu@Ag/CNFs).

The metal precursors were quickly reduced due to the apparent higher redox potential of H₂PO₂⁻¹ than Cu²⁺. During the reaction process, the blue copper sulphate solution undergoes decolourization, transitioning to a colourless state, followed by a shift to green, and ultimately resulting in a wine-red colour, indicating the formation of copper nanoparticles. In the subsequent step, a silver shell is formed on the surface of the copper core via a transmetallation reaction. In this process, silver salt is introduced, and the copper metal facilitates the reduction of silver ions at the surface of the Cu NPs, leading to the deposition of a silver shell on the copper core, resulting in the formation of the Cu@Ag/CNFs.

This approach is especially helpful for the preparation of core–shell nanostructures because of the significant difference between the reduction potentials of noble metals like Ag and non-noble metals like Cu. Here, the silver, which has a higher reduction potential, is reduced by the surface of the core Cu nanoparticles, forming a Cu@Ag core–shell structure on the surface of the CNFs. The corresponding chemical equations are as follows:

Cu²⁺ + 2e⁻ → Cu⁰ E° = +0.34 V (2)

Ag⁺ + e⁻ → Ag⁰ E° = +0.80 V (3)

Net reaction: Cu⁰ + 2Ag⁺ → Cu²⁺ + 2Ag⁰ +1.05 V (4)

The as-synthesized copper nanoparticles served as the core particles for the direct deposition of a uniform silver shell on their surfaces, resulting in the formation of core–shell nanostructures. Here, the exposed Cu surface acts as a reducing agent for the silver ions (Ag⁺) to get reduced into silver nanoparticles (Ag⁰). The reduction of silver ions (Ag⁺) by Cu metal occurs spontaneously, as the thermodynamic conditions favor the overall reaction, eliminating the necessity for any external reducing agents in the reduction process.

Structural and morphological characterization

The structural characterization of the developed CNFs, Cu@Ag/CNFs (Cu : Ag =1 : 1, 1 : 3, 1 : 5) was carried out by analyzing the X-ray diffractograms, as shown in Fig. 1. The XRD analysis shows the appearance of a broad diffraction peak at 25.6° for the CNFs and Cu@Ag/CNFs, which is attributed to the amorphous phase of the graphitic carbon ((002) plane) present in the carbon nanofibers associated with hexagonal graphite (JCPDS card 41–1487).²⁴ The crystalline peaks centred at 2θ ≈ 38.1°, 44.3°, 64.5°, 77.4° and 82.5° correspond to the (111), (200), (220), and (311) crystal planes of the face-centred cubic structure of Ag (JCPDS file No. 04-0783). The (200) and (311) planes of Ag were found to be overlapped with the (111) and (220) planes of Cu, respectively. Further evidence indicates that all copper atoms are effectively shielded and converted into Cu@Ag core–shell nanoparticles as demonstrated by the lack of any copper peaks in the X-ray diffraction pattern.

Fig. 1 XRD analysis of carbon nanofibers (black), and Cu@Ag/CNFs with different ratios of Cu : Ag = 1 : 1 (red), 1 : 3 (blue), 1 : 5 (pink).

The absence of diffraction peaks corresponding to the copper core atoms suggests that these copper atoms are in a kinematic diffraction state, resulting in the observation of only the silver diffraction peaks.^25,26 Furthermore, the formation of core–shell structures rather than alloys and mixtures was also aided by the significant reduction potential difference between these two metals (0.46 eV) and the difference in the lattice constants (0.048 nm) between copper and silver.^27,28 Hence, the XRD analysis demonstrated that the Cu–Ag core–shell nanoparticles were exclusively composed of copper (Cu) and silver (Ag), with no detectable impurities.

The absence of additional crystalline phases confirms the successful formation of Cu@Ag/CNFs. Also, the XRD analysis demonstrated a correlation between the increasing Ag ratio and the enhancement of the Ag peak intensities, indicating the successful formation of silver shells on the surface of the copper nanoparticles. Furthermore, an elevation in Ag⁺ ion concentration results in a corresponding increase in the intensity of silver peaks in the XRD pattern, which can be attributed to the agglomeration of silver nanoparticles. The Scherrer equation was used to determine the crystallite size of the Cu@Ag nanoparticles:

D = Kλ/β cos θ, (5)

where d is the average crystallite size of the nanoparticles, λ is the X-ray wavelength (Cu Kα) = 0.154, β is the full width half maximum (FWHM) of the peaks in radians, θ is the angle of diffraction and K is the Scherrer constant, which is equal to 0.89. The average crystallite size for the silver and copper nanoparticles calculated from the Ag (111) and Cu (111) peaks using the Scherrer equation, was found to be 11.2 nm and 10.7 nm, respectively. The surface morphology of the core@shell Cu@Ag nanoparticles anchored on the surface of the carbon nanofibers was examined using scanning electron microscopy as shown in ESI,† S1.

Transmission electron microscopy analysis was used to investigate the precise morphology, structure, and distribution of the core@shell Cu@Ag anchored on CNFs (Fig. 2). The continuous fibrous structure of carbon nanofibers with a diameter of approximately 200–300 nm was confirmed by TEM analysis, as depicted in Fig. 2(a). The lower magnification images show the uniform dispersion of Cu@Ag nanoparticles, which were found to be firmly bound to the surface of the carbon nanofibers (Fig. 2(a and b)). The core@shell structure of the Cu@Ag bimetallic nanoparticles anchored on the CNFs has been further confirmed from the HR-TEM image displayed in Fig. 2(c), where the core copper nanoparticles are visible with a darker contrast surrounded by a shell layer of silver. This confirms the successful formation of core Cu nanoparticles surrounded by a uniform layer of silver shell, anchored on the surface of carbon nanofibers. The HRTEM analysis is further supported by the selected area diffraction (SAED) pattern analysis shown in Fig. 2(d), which reveals the spot patterns assigned to the (111), (200), (220), (311), and (222) planes of fcc Ag crystals and the (111) and (220) planes of fcc Cu crystals. These patterns demonstrate the polycrystalline nature of Ag and Cu, and they are in good agreement with the XRD analysis of the core–shell Cu@Ag/CNFs, which demonstrates the successful formation of bimetallic Cu@Ag core–shell nanoparticles tethered on carbon nanofibers.

Fig. 2 (a) and (b) Representative TEM images (scale bar A = 200 nm and B = 100 nm), (c) HR-TEM image of Cu@Ag core shell nanoparticles (scale bar = 5 nm), (d) corresponding SAED pattern of Cu@Ag/CNFs (scale bar = 51/nm), (e) HRTEM images of and Cu@Ag/CNFs (scale bar = 5 nm). Inset shows IFFT profiles with a lattice d-spacing value of 0.23 nm consistent with the (111) crystalline plane of fcc Ag and the lattice spacing of 0.20 nm, analogous to the (111) crystalline plane of the fcc structure of Cu on Cu@Ag/CNFs and (f) TEM-EDS analysis of Cu@Ag/CNFs confirming the presence of carbon, copper and silver.

Fourier transform (FFT) analysis was used to investigate the high-resolution images of Cu@Ag/CNFs. The Fourier transform was imaged in real space by masking off the desired frequencies to eliminate unwanted noise. In order to obtain the real image with sub-angstrom resolution, an inverse Fourier transform (IFFT) was performed on the Fourier masked image, as shown in Fig. 2(e). The lattice spacing of Cu was found to be 0.23 nm, which is analogous to the (111) crystalline plane of the fcc structure of Ag shown in Fig. 2(e). On the other hand, the inverse fast Fourier transform analysis from the HRTEM images of the Cu@Ag/CNFs yields a lattice d-spacing value of 0.20 nm, which is consistent with the (111) crystalline plane of fcc Cu on the Cu@Ag/CNFs. The JCPDS database indicates that Ag and Cu nanoparticles exist as bimetallic nanoparticles in the Cu@Ag/CNFs, which is consistent with the reported lattice spacing of 0.23 nm and 0.20 nm corresponding to the (111) and (111) crystal planes of fcc Ag and fcc Cu. Fig. 2(f) shows the EDS analysis of the Cu@Ag/CNFs, which confirms the uniform decoration and successful formation of the core@shell Cu@Ag nanoparticles on the surface of the carbon nanofibers. The detailed investigation of the Cu@Ag/CNFs using TEM analysis reveals that the carbon nanofibres had a high surface area to volume ratio, which allowed for better binding with the analyte molecules and Cu@Ag nanoparticles were evenly deposited on the smooth surface of the CNFs. This allowed for the successful anchoring of core–shell Cu@Ag nanoparticles with multiple hotspots to produce high enhancement contributing to SERS. The XRD analysis and TEM analysis of the Cu@Ag/CNFs were consistent with one another, indicating the integration and successful formation of Cu@Ag/CNFs.

Evaluation of the SERS enhancement properties of the Cu@Ag/CNFs

The evaluation of the performance of the Cu@Ag/CNF-based sensing material to be used as a SERS substrate was initially evaluated using the common Raman reporter molecule, rhodamine 6G (R6G). A confocal Raman microscope equipped with a 532 nm laser was used to acquire the SERS spectra. At the sample position, the laser power was set as 1 mW. To ensure reproducibility, the spectra were randomly acquired at three different locations, with 10 s accumulations and 20 points being recorded. With individual concentrations of 10⁻¹ M and 10⁻⁶ M, respectively, the normal Raman spectrum and the SERS spectrum of the R6G molecules were obtained using glass substrates coated with and without the SERS active material, Cu@Ag/CNFs. In the presence of the Cu@Ag/CNF-based SERS substrate, all of the distinct bands corresponding to R6G were found to be greatly enhanced, but only low intensity peaks of R6G were seen in the normal Raman spectra in the absence of the SERS active material, as depicted in Fig. 3(a). The characteristic bands of R6G are attributed to the C–C–C ring in-plane bending, the C–H out-of-plane bending, the C–C stretching mode, and the C–C stretching in xanthene, respectively. The corresponding bands appeared at 615 cm⁻¹, 775 cm⁻¹, 1188 cm⁻¹, 1315 cm⁻¹, 1365 cm⁻¹, 1510 cm⁻¹, 1574 cm⁻¹, and 1645 cm⁻¹. Additionally, the other peaks that appeared at 1365, 1510, and 1645 cm⁻¹ are caused by the aromatic C–C stretching vibrations, which is consistent with earlier reports.^29,30 This massive enhancement results from the strong localization of the near field intensity, creating hot-spots where the Raman signal of the analyte molecules can be amplified by several orders of magnitude. The smooth cylindrical structures of the carbon nanofibers have a greater surface area, which serves as an open interior to accommodate Cu@Ag nanoparticles closely together. Furthermore, the Cu@Ag/CNFs can retain the probe molecules together in their close proximity. Therefore, the hotspot generation between Cu@Ag nanoparticles and the increase in the number of probe molecules adsorbed on the Cu@Ag/CNFs results in enhanced Raman signals. The formation of these so-called hot spots between nearly contacted nanoparticles encourages the excitation of abnormally high electromagnetic fields, which effectively intensifies the Raman signals contributing towards SERS. The lack of any distinctive peaks of R6G in the normal Raman spectra is due to the lack of SERS active material, which eliminates the possibility of a strong chemical and electromagnetic interaction between the analyte and Cu@Ag/CNFs.

Fig. 3 (a) Comparison of normal Raman (0.1 M) and SERS spectra (10 μM) of R6G using Cu@Ag/CNFs based SERS substrates. Comparison of the SERS spectra of R6G at (b) different ratios of Cu : Ag (1 : 1 (black), 1 : 3 (blue), 1 : 5 (red)) and (c) different concentrations ranging from 10⁻⁶ to 10⁻¹² M, and (d) effect of concentration as a function of peak intensity at 615 cm⁻¹.

The intensity of the appropriate R6G peak (1365 cm⁻¹) measured in the SERS spectra was compared to the equivalent peak measured from an aqueous solution of R6G in normal Raman spectra to calculate the SERS enhancement factor values. To obtain accurate results, baseline correction has been applied to the SERS spectra of R6G before the enhancement factor was calculated. The equation used for calculating the substrate enhancement factor per molecule is given by:

Enhancement factor = I_SERS × N_Raman/I_Raman × N_SERS (6)

where, N_SERS and N_Raman denote the number of molecules probed in the laser spot on the SERS substrates and in the aqueous sample, respectively and I_SERS and I_Raman represent the corresponding normal SERS and Raman intensities, respectively. The enhancement factor was calculated using the normal Raman spectra of 0.1 M R6G, whereas SERS spectra were recorded at concentrations in the range of micro molars (10 μM). The number of molecules sampled in the bulk R6G was calculated using a laser spot size of 2 μm (1 μm radius), a penetration depth of 10 μm, and the numerical aperture of the long working distance objective (50 × LWD, N_A = 0.5). The probed volume was approximated as a cylinder with a height of 10 μm (the laser's penetration depth) and a diameter of 2 μm (1 μm radius). The following equation was used to determine the number of molecules being probed:

Number of molecules = πr²h_cN_A (7)

The volume in the laser spot size, πr²h (volume of cylinder) was calculated as 3142 μm³. Based on the above-mentioned values, the number of molecules in the bulk (N_Raman) was determined to be approximately 1.89 × 10¹⁴, while the number of molecules probed on the surface (N_SERS) was estimated to be approximately as 1.89 × 10⁶. The intensities of the peak at 1365 cm⁻¹ for SERS and Raman were determined to be 12 327 and 297, respectively, based on the obtained Raman and SERS spectra. The ratio of intensities was calculated as I_SERS/I_Raman = 12 327/297. By substituting this ratio in eqn (6), the enhancement factor was calculated as ∼4.1 × 10⁷, demonstrating the efficiency of the Cu@Ag/CNF based SERS substrate to detect a trace level concentration of analyte molecules.

In order to evaluate the SERS performance of the developed material, different core-shell ratios of Cu and Ag, such as 1 : 1, 1 : 3, and 1 : 5, were prepared, and the SERS performances were studied using R6G, as shown in Fig. 3(b). The comparison of the SERS performances of R6G using different ratios of Cu : Ag revealed that the maximum enhancement was achieved for the Cu : Ag ratio of 1 : 3 showing an intensity enhancement up to 20 000 counts. This could be explained in terms of the agglomeration effect and the risk of oxidation of the copper nanoparticles.³¹

Following the transmetallation reaction, which produces core@shell Cu@Ag NPs, the Ag content on the surface of the Cu nanoparticles increases than in the vicinity of the nanoparticles. Consequently, for the molar ratio of Cu : Ag = 1 : 5, an excess of silver in the Cu@Ag core–shell nanoparticle contributes to agglomeration resulting in a lower SERS enhancement. Due to this agglomeration effect, the silver nanoparticles will get agglomerated and get deposited away from the surface of the carbon nanofibers limiting the number of hotspots generated and thereby the SERS enhancement. This indicates that the Cu@Ag core–shell nanoparticles are less agglomerated as the molar ratio of [Cu]/[Ag] decreases, which in-turn results in higher SERS enhancement. For the molar ratio of Cu : Ag as 1 : 3, the surface of Cu was found to be uniformly deposited with Ag, resulting in the formation of a typical copper core with a silver shell layer. This assists in increasing the number of hot spots between the Cu@Ag nanoparticles, ultimately improving the SERS performance. But in the case of the Cu : Ag molar ratio of 1 : 1, the SERS enhancement was found to be decreased. This could be due to the fact that the silver precursor is not sufficient to completely cover the surface of the copper core particles and Ag appears only partially on the surface of the Cu nanoparticles, and is therefore exposed to oxidation. This may be the probable reason for the lower enhancement in the case of Cu : Ag = 1 : 1. The corresponding TEM images of the Cu@Ag/CNFs at different Cu : Ag ratios of 1 : 1, 1 : 3 and 1 : 5 are given in the ESI,† Fig. S2. Thus, it can be concluded from the above investigation that the molar ratio strongly influences the structure and the extent of coverage of Ag on the surface of the Cu nanoparticles and thereby the SERS enhancement. Hence, better SERS performance and high enhancement were achieved for the ideal Cu : Ag molar ratio of 1 : 3 due to the formation of the optimal Cu@Ag core–shell nanostructures on the CNFs. At this optimal ratio, two highly opposing features: (1) agglomeration associated with excess Ag, and (2) the risk of oxidation of the copper core due to low supply of silver can be minimized.

SERS spectra were recorded for various concentrations of R6G ranging from 10⁻⁶ M to 10⁻¹² M in order to determine the detection limit using the developed substrate, as depicted in Fig. 3(c). The results revealed that the primary characteristic peaks of the probe molecules are apparent even at concentrations down to 10⁻¹² M. The SERS peak intensities were found to be decreased with decrease in the concentration of probe molecules ranging from 10⁻⁶ to 10⁻¹² M, as expected. The relationship between concentration and peak intensity at 615 cm⁻¹ is shown in Fig. 3(d), which gives a linear trend with an R² value of 0.9931. The above mentioned observations clearly highlight that the suggested Cu@Ag/CNFs based SERS substrate has the lowest LOD of 10⁻¹² M and the highest EF in the order of 10⁷, demonstrating that the method offers a sensitive protocol for R6G detection that can identify the probe molecule even at concentrations lower than 10⁻¹² M. This demonstrates the effectiveness of the developed Cu@Ag/CNF-based SERS substrate in detecting analytes at extremely low concentrations, expanding its applications towards trace level detection and real-time analysis of saliva samples.

Morphology, composition, and dielectric environment of the metal nanostructures generally determine the plasmonic effects responsible for the SERS features. In addition, in a matrix of multiple nanoparticles, the coupling of inherent plasmon oscillations of adjacent nanoparticles needs to be addressed in detail while evaluating such electromagnetic effects. However, quantitative measurements on the plasmonic effects at the interparticle junctions (typically known as ‘hotspots’) using conventional experimental methods are highly challenging. Alternatively, researchers have often performed theoretical simulations to precisely and rapidly monitor such hotspots. In the present nanosystem, as shown in Fig. 4(a), many Cu@Ag core–shell nanostructures are located near each other on the surface of carbon nanofibers; therefore, it is possible to couple the plasmon oscillations of adjacent nanostructures. To estimate the extent of field-intensity enhancement around the completely isolated and physically adjacent nanostructures, the simulations were carried out using commercially available Ansys Lumerical FDTD solution software. During these simulations, two different configurations (type 1 and type 2) were modeled with realistic conditions to obtain the most accurate results, as shown in Fig. 4(b). In the type 1 configuration, an isolated nanostructure was chosen to evaluate the plasmonic responses.

Fig. 4 (a) Cartoon depict of Cu@Ag core–shell nanoparticle-anchored carbon nanofibers, (b) the simulation structures and respective electric-field intensity enhancement profiles, and (c) the field-intensity line profiles. The dashed lines in (b) indicate the specific direction in which the field-intensity is measured.

On the other hand, in type 2, a dimer configuration with an interparticle gap of 2 nm between the individual nanostructures was modeled. A light source with a wavelength of 532 nm was used to illuminate the simulation structure from the topside. Upon light interaction, the isolated nanostructure generated near-field enhancement mostly on its lateral sides i.e., perpendicular to the propagation direction of light. The line profile A (a white dashed line in Fig. 4(b)) indicates the variation of the field intensity enhancement factor (|E|²/|E₀|²) with distance. A maximum |E|²/|E₀|² (∼14) was observed close to the surface of the nanostructure and it continuously decreased as the distance from the surface increased. In the case of the dimer configuration, a plasmonic hotspot was observed at the interparticle junction indicated by a rectangle in Fig. 4(b). The line profiles B and C drawn through the hotspot clearly indicate the variation of field strength at the hotspot. A high |E|²/|E₀|² value more than 100 was observed across the hotspot region with a maximum value of ∼252 at the center, which is shown by the sharp peaks in the line profiles B and C. These findings reveal that the hotspots induced via the coupling of adjacent nanostructures have the potential to generate a strong field intensity across a wide region, which is indeed beneficial for carrying a large number of analytes to attain huge enhancement in Raman signals of targeted analytes.

SERS detection of nitrate, nitrite and thiocyanate

Oral cancer is the world's sixth most common cancer and continues to be the leading cause of oral disease-related death worldwide. The oral mucous membrane can undergo a range of changes as a result of tobacco use which can cause alteration in the composition of saliva. Most important biomarkers in saliva towards oral cancer include nitrate, nitrite and thiocyanate. Drinking water, fruits, vegetables, and salty sausages containing nitrite ions enter the digestive tract through the oral cavity and ultimately reach the salivary glands through the blood circulation.^32–34 Though nitrates are not inherently toxic, they can catalyze reactions with other substances, potentially leading to the formation of toxic compounds. Many studies have shown that patients with oral carcinoma exhibit significantly elevated levels of salivary nitrates and nitrites compared to control groups.³⁵ This is because the saliva of habitual tobacco chewers and smokers contains nitrosamines, which are likely leached from the tobacco or formed in situ through the nitrosation of tobacco alkaloids as well as that of secondary amines.³⁶ It also implies that chewers of tobacco might be subjected to ongoing exposure through their oral mucosa to nitrosamines and other nitroso compounds. In the control group, salivary nitrite levels range from 30 to 100 μg L⁻¹, while multiple studies have shown that patients with oral cancer exhibit significantly elevated nitrite levels (>200 μg L⁻¹).³⁷ Thiocyanate is another significant biomarker for diagnosing oral cancer, which is a naturally occurring component found in body fluids and enters the body through the intake of Brassica vegetables, milk, and cheese. It is one of the important cyanide metabolic by-products and a major marker of smoke inhalation since the tobacco smoke gas mainly consists of carbon monoxide, and hydrogen cyanide. Elevated SCN-ion levels in smokers' saliva increase their risk of developing cancer by causing the nitrosation process, which turns organic compounds into nitrosoderivatives, strong teratogens and carcinogens.³⁸ Salivary SCN-concentrations in healthy non-smokers range from 0.5 to 2 mM, with an average of 1 mM. However, salivary concentrations of thiocyanate in heavy smokers can reach up to 6 mM.³⁹ Hence, the detection of these important biomarkers in saliva can be used as a detection tool for the pre-diagnosis of oral carcinoma.

The viability of SERS for the detection of these biomarkers (nitrate, nitrite, and thiocyanate) was first investigated in order to determine the effectiveness of Cu@Ag/CNFs for clinical applications. Using the Cu@Ag/CNF-based SERS substrate, the Raman spectra (0.1 M) and the SERS (100 μM) corresponding to the nitrate, nitrite and thiocyanate were compared, which is depicted in Fig. 5. As predicted the SERS spectra showed clearly identifiable peaks of nitrate, nitrite, and thiocyanate with good intensity enhancement compared to normal Raman spectra. The SERS spectra of nitrate (Fig. 5a) showed a sharp peak centred at 1044 cm⁻¹, resulting from the symmetric stretching mode, v1 of nitrate, which is in good agreement with reported values.^40–42Fig. 5b shows the comparison of the Raman and SERS spectra of nitrite in which the SERS spectra showed a clearly distinguishable peak of nitrite centred at around 1330 cm⁻¹, which is caused by the symmetric N–O stretching vibrations of the nitrite.^43,44

Fig. 5 Comparison of Raman (0.1 M) and SERS spectra (100 μM) of (a) nitrate, (b) nitrite and (c) thiocyanate using Cu@Ag/CNF-based SERS substrates.

The SERS spectra of thiocyanate shown in Fig. 5c encounter two prominent peaks that are centred at 747 cm⁻¹ and 2150 cm⁻¹, respectively. These peaks indicate SCN and were in good agreement with earlier reports.^45,46 The asymmetric stretching vibration of the –C≡N bond is responsible for the peak at 2150 cm⁻¹, which is selected as the characteristic peak for identifying SCN⁻ due to its high intensity. The absorption peak at 753 cm⁻¹ is caused by the stretching vibration of the C–S bond. In the SERS spectra, the highest Raman peak, located at 2150 cm⁻¹, displayed an intensity enhancement of about 12 000 counts. Hence, the comparison of Raman and SERS spectra showed strong intense SERS peaks of nitrate, nitrite and thiocyanate with good intensity enhancement. This proves the efficacy of the SERS active Cu@Ag/CNF-based SERS substrate to detect a very low concentration of these biomarkers, especially in complex mixtures such as saliva.

As a proof-of-concept investigation, the potential of the Cu@Ag/CNF based SERS sensor platform for the detection of cancer biomarkers was demonstrated by employing real time saliva sample analysis. Using a 532 nm laser, the SERS spectra of saliva samples of 20 confirmed cases of oral squamous cell carcinoma patients and 24 control specimens were analysed. Using the pre-optimized instrument settings and experimental conditions, the Raman spectral analysis was performed on these salivary samples. The SERS spectra were obtained in the range of 775 cm⁻¹ and 2250 cm⁻¹ for a detailed investigation, as all significant characteristic peaks corresponding to the salivary biomarkers were discovered to be located within this spectral range. Since saliva is a very complex mixture, several background characteristics from saliva were evident in the SERS spectra, which may complicate the spectral analysis as some of them overlap with analyte spectral features. Hence, the collected data were further processed by removing the background trace and subtracting fluorescence. The results of the spectral analysis, shown in Fig. 6, compare the SERS spectra of healthy controls and patients with oral cancer.

Fig. 6 Comparison of the SERS spectra of the saliva of oral cancer patients (black) and healthy controls (red).

The SERS spectra of the control group are marked as a red line and the black line represents the SERS spectra of oral cancer saliva specimens. These spectra demonstrated several significant spectral peaks that varied between the healthy and cancer groups. The SERS analysis revealed that all of the peaks in the SERS spectra that corresponded to nitrate (1049 cm⁻¹), nitrite (816 cm⁻¹ and 1330 cm⁻¹), and thiocyanate (747 cm⁻¹ and 2150 cm⁻¹) were present and were highly distinguishable. Compared to the healthy controls, it was discovered that the oral cancer groups had significantly up-regulated levels of these three important biomarkers. Apart from this, many other peaks were also enhanced in the SERS spectra, which were found to be upregulated/downregulated in the oral cancer saliva groups compared to healthy controls (Table 1).

Table 1 Raman Shift and the corresponding assignment of peaks in the saliva of oral cancer patients and healthy control group. (Red represents the peaks which are down-regulated in oral cancer patients’ saliva)

Raman shift (cm^−1)	Assignment
747	Thiocyanate
816	Nitrite
885	Cellobiose
1049	Nitrate
1126	Sucrose
1224	Amide III
1275	Phospholipids
1330	Nitrite
1417	Nucleic acids amide III
1543
1659
2150	Thiocyanate

---

In the oral cancer group, the upregulated peaks include amide III (β sheet structure, 1224 cm⁻¹); amide III (1275 cm⁻¹); υ^s COO–IgG, 1409 cm⁻¹; and C=C stretching in the quinoid ring (1417 cm⁻¹). Peaks related to the disaccharide (cellobiose) (C–O–C) skeletal mode (885 cm⁻¹), n(C–C) skeletal mode of the acyl backbone in lipids (transconformation), C–N stretching vibration (protein vibration), v(C–O) + v(C–C), disaccharides, and sucrose (1126 cm⁻¹) were less evident and found to be down-regulated in cancer specimens. For samples from cancer patients, some peaks that appeared at 1417 cm⁻¹ in saliva and 1543 cm⁻¹ in oral cells are indicative of C–H bending and COO– stretching vibrations. All the characteristic peaks were in good agreement with the previous reports, demonstrating that it can be used as a diagnostic tool to distinguish between the oral cancer groups and healthy controls.⁴⁷

3D printed microfluidic platform for multiple sample analysis

A 3D printed microfluidic platform was used for the analysis of the saliva samples from both healthy controls and oral cancer groups. The microfluidic platform was designed with a main well and 12 sub-wells connected through channels. This microfluidic sensor platform can be used to analyse the reproducibility and repeatability of the SERS analysis (Fig. 7a). The chip had overall dimensions of 4 cm diameter and a height of 0.5 cm (thickness). The main well consists of 1 cm diameter and 0.2 cm depth well for the saliva sample collection. It is connected to the sensor arrays (sub wells) using channels, which are at an angle of 10° with length 0.6 cm. The sub wells/sensor array has a dimension of 0.2 cm diameter and 0.1 cm depth for the collection of saliva samples. The entire device is designed in such a way that when the saliva is dropped on the main well, it will flow (laminar flow) through the channels to the sensor array where the saliva can be collected in the 12 wells (Fig. 7b), which will be analysed as shown in the ESI,† S3. This device can be used to evaluate the reproducibility of the SERS spectra in which a single saliva sample can give rise to 12 spectra and hence the number of data acquired will be more.

Fig. 7 (a) Design of the microfluidic SERS sensor platform designed using SOLIDWORKS 3D CAD design software and (b) a 3D printed SERS microfluidic platform.

AI approach using PCA-random forest analysis

The Raw SERS spectra obtained were processed by vector normalization, smoothing, and baseline correction in order to avoid errors. The variations in the Raman peaks of the oral cancer and healthy control groups are analyzed in Fig. 8a, which compares the average SERS and standard deviation (SD) of the oral cancer group (n₁ = 56) with the control group (n₂ = 60). The red and green lines represent the average SERS spectra of the saliva of oral cancer patients and healthy controls, respectively. The dotted line represents the standard deviations, whereas the blue line represents the difference between the two spectra.

Fig. 8 (a) Comparison of the average spectra for oral cancer patients (red line, n = 56) versus that of the normal group (green line, n = 60). The dotted line represents the standard deviations of the averages. The blue line shown at the bottom is the difference in the spectra. (b) Scatter plot of scores from 2 PCs (PC-1 and PC-2) for healthy volunteers (blue) vs. oral cancer patients (red) and (c) 3D scatter plot using PC-1, PC-2 and PC-3 as the 3 axes for healthy volunteers (green) vs. oral cancer patients (red) and (d) bar chart showing the accuracy, specificity, sensitivity and F measure using PCA combined with the random forest algorithm.

To demonstrate the performance of the proposed similarity model, the data set was divided into two parts; 80% of all ratings were selected as the training data set and the remaining 20% of the ratings were considered as the test data. For measuring classification accuracy, k (= 5) cross validation was done by randomly choosing different train and test data sets. In this study, principal component analysis (PCA) combined with random forest was used as a powerful approach for the classification of healthy and oral cancer patients. PCA was adopted to reduce the high-dimensional SERS data into a smaller set of principal components, which capture the most significant variance in the data. This dimensionality reduction simplifies the dataset, making it more manageable and reducing the risk of overfitting, ensuring that only the most relevant information is retained.

PCA with the random forest algorithm was utilized for classification, capitalizing on its ability to handle complex interactions between features and its robustness against overfitting. Here, the algorithm for the dimensions from 10–100 in steps of 10 was executed and it was found that 20 was the optimal dimension and the experimental results were subjected to PCA with 20 dimensions. In order to make use of the data acquired from the entire spectrum, PCA-RF techniques were employed to evaluate and distinguish saliva SERS spectra between oral cancer patients and the healthy volunteers. Scatter plots comparing oral cancer patients with healthy controls were developed based on different combinations of significant PCs. Fig. 8(b) displays the data distribution according to PCA with two dimensions (PC-1 and PC-2), where PC-1 is represented by the horizontal axis and PC-2 by the vertical axis.

Fig. 8(c) displays a three-dimensional scatter plot with PC-1, PC-2, and PC-3 as the three axes. The slanting plane represents the 3D plane separating the healthy and oral components. Research studies expose various mathematical measures to evaluate the performance of classification algorithms. In this investigation, measures namely accuracy, specificity, sensitivity and F-measure were used to evaluate the performance of the classification algorithm. Accuracy is the fraction of cases the model correctly predicted, sensitivity (recall) is defined as the fraction of positive cases predicted as positive, specificity is defined as the fraction of negative cases predicted as negative and precision is the fraction of true positive cases from all the cases the model predicted positive. F measure is the harmonic mean of precision (P) and recall (R). Table 2 depicts the corresponding mathematical formulae, where TP, TN stand for true positives and true negatives respectively and FP, FN stand for false positives and false negatives, respectively.

Table 2 Classification performance measures used in PCA-RF analysis

---

The findings of this study show that the accuracy, specificity, sensitivity, and F-measure are 87.5%, 92%, 88%, and 88.5% respectively, as shown in Fig. 8(d). This indicates that the proposed PCA with the RF method will correctly identify cancerous patients approximately 88% of the time. Additionally, the test correctly classified 87.5% of the individuals, indicating excellent differentiation between the cancer and control groups. The findings of this study provide solid proof for the use of SERS spectral and multivariate analysis as a promising clinical adjunct in distinguishing between healthy individuals and oral cancer patients.

Conclusions

The integration of a low-cost, label-free surface enhanced Raman spectroscopy (SERS) biosensor within a microfluidic platform represents a notable breakthrough in oral cancer biomarker detection. By utilizing bimetallic copper@silver (Cu@Ag) core–shell nanoparticles anchored on carbon nanofibers, this biosensor achieves exceptional signal enhancement up to 10⁷ fold with a detection limit as low as 10⁻¹² M with rhodamine 6G enabling highly sensitive and reproducible detection of biomarkers in saliva samples. Integrating a microfluidic platform for high-throughput, multiplexed sample analysis further improves reliability, while advanced AI approaches, including principal component analysis and random forest classification (PCA-LDA), enhance diagnostic accuracy achieving an overall classification accuracy of 87.5%, along with specificity of 92% and sensitivity of 88%. This microfluidic platform integrated SERS biosensor platform stands out as a pioneering tool in non-invasive oral cancer diagnostics, highlighting the importance of interdisciplinary innovation in advancing cancer detection technologies.

Ethical approval

All procedures performed in this study were in accordance with the ethical standards. Ethical clearance was obtained from the Institutional Human Ethics Committee, PSG Institute of Medical Science and Research, Coimbatore (Ref No: PSG/IHEC/2023Appr/FB/014) dated 23.02.2023 for the collection of human saliva samples.

Author contributions

The manuscript was written through contributions from all authors. All authors have given approval to the final version of the manuscript.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request. All relevant datasets, including raw and processed data from SERS measurements, machine learning model outputs, and fabrication protocols for the microfluidic platform and SERS sensors, are stored securely and can be accessed for research purposes. The data supporting this article have been included as part of the ESI.†

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work is funded by a DST-INSPIRE Fellowship (No. DST/INSPIRE Fellowship/2020/IF200094), and DST-WTI grant (DST/TMD-EWO/WTI/2K19/EWFH/2019/273) Government of India, Ministry of Science and Technology, Department of Science and Technology. The authors wish to acknowledge the Department of Science and Technology, Government of India for the financial support from grants, DST-INSPIRE (DST/INSPIRE Fellowship/2020/IF200094) and DST-WTI (DST/TMD-EWO/WTI/2K19/EWFH/2019/273). The authors also acknowledge the facilities and support provided by the management, PSG Institute of Advanced Studies, Coimbatore.

References

1. N. Gupta, R. Gupta, A. K. Acharya, B. Patthi, V. Goud, S. Reddy, A. Garg and A. Singla, Changing Trends in oral cancer-a global scenario, Nepal J. Epidemiol., 2016, 6(4), 613.
2. M. Guillon, N. P. Dang, J. Thévenon and L. Devoize, Salivary diagnosis of oral cancers by salivary samples: a systematic literature review, J. Oral Med. Oral Surg., 2021, 27(3), 39.
3. A. Zalewska, N. Waszkiewicz and R. M. López-Pintor, The use of saliva in the diagnosis of oral and systemic diseases, Dis. Markers, 2019, 2019.
4. M. Hardy, L. Kelleher, P. de Carvalho Gomes, E. Buchan, H. O. Chu and P. Goldberg Oppenheimer, Methods in Raman spectroscopy for saliva studies–a review, Appl. Spectrosc. Rev., 2022, 57(3), 177–233.
5. E. Brindha, R. Rajasekaran, P. Aruna, D. Koteeswaran and S. Ganesan, High wavenumber Raman spectroscopy in the characterization of urinary metabolites of normal subjects, oral premalignant and malignant patients, Spectrochim. Acta, Part A, 2017, 171, 52–59.
6. A. Falamas, C. I. Faur, S. Ciupe, M. Chirila, H. Rotaru, M. Hedesiu and S. C. Pinzaru, Rapid and noninvasive diagnosis of oral and oropharyngeal cancer based on micro-Raman and FT-IR spectra of saliva, Spectrochim. Acta, Part A, 2021, 252, 119477.
7. K. Delli, F. K. Spijkervet, F. G. Kroese, H. Bootsma and A. Vissink, Xerostomia, Saliva: Secretion and functions, 2014, vol. 24, pp. 109–125.
8. S. Campuzano, P. Yánez-Sedeño and J. M. Pingarrón, Electrochemical bioaffinity sensors for salivary biomarkers detection, TrAC, Trends Anal. Chem., 2017, 86, 14–24.
9. I. T. Gug, M. Tertis, O. Hosu and C. Cristea, Salivary biomarkers detection: Analytical and immunological methods overview, TrAC, Trends Anal. Chem., 2019, 113, 301–316.
10. Y. Kalachyova, O. Guselnikova, R. Elashnikov, I. Panov, J. Z̆ádny, V. Církva, J. Storch, J. Sykora, K. Zaruba, V. Švorčík and O. Lyutakov, Helicene-SPP-based chiral plasmonic hybrid structure: toward direct enantiomers SERS discrimination, ACS Appl. Mater. Interfaces, 2018, 11(1), 1555–1562.
11. H. Shi, H. Wang, X. Meng, R. Chen, Y. Zhang, Y. Su and Y. He, Setting up a surface-enhanced Raman scattering database for artificial-intelligence-based label-free discrimination of tumor suppressor genes, Anal. Chem., 2018, 90(24), 14216.
12. C. Danciu, A. Falamas, C. Dehelean, C. Soica, H. Radeke, L. Barbu-Tudoran, F. Bojin, S. C. Pînzaru and M. F. Munteanu, A characterization of four B16 murine melanoma cell sublines molecular fingerprint and proliferation behavior, Cancer Cell Int., 2013, 13, 1–2.
13. V. K. Rao and T. P. Radhakrishnan, Tuning the SERS response with Ag-Au nanoparticle-embedded polymer thin film substrates, ACS Appl. Mater. Interfaces, 2015, 7(23), 12767–12773.
14. M. Y. Khaywah, S. Jradi, G. Louarn, Y. Lacroute, J. Toufaily, T. Hamieh and P. M. Adam, Ultrastable, uniform, reproducible, and highly sensitive bimetallic nanoparticles as reliable large scale SERS substrates, J. Phys. Chem. C, 2015, 119(46), 26091–26100.
15. S. Tian, W. You, Y. Shen, X. Gu, M. Ge, S. Ahmadi, S. Ahmad and H. B. Kraatz, Facile synthesis of silver-rich Au/Ag bimetallic nanoparticles with highly active SERS properties, New J. Chem., 2019, 43(37), 14772–14780.
16. F. Lin, M. Fan, Y. Liao, X. Wei, J. Huang, T. Zhou and Y. Lu, Synthesis of Cu–Ag core–shell nanoparticles as highly active surface-enhanced Raman scattering substrates for sensitive detection of caffeic acid, Mater. Express, 2020, 10(5), 687–693.
17. M. Cazayous, C. Langlois, T. Oikawa, C. Ricolleau and A. Sacuto, Cu-Ag core-shell nanoparticles: A direct correlation between micro-Raman and electron microscopy, Phys. Rev. B: Condens. Matter Mater. Phys., 2006, 73(11), 113402.
18. X. Jin, A. Mao, M. Ding, P. Ding, T. Zhang, X. Gu, W. Xiao and J. Yuan, A simple route to synthesize Cu@ Ag Core–shell bimetallic nanoparticles and their surface-enhanced Raman scattering properties, Appl. Spectrosc., 2016, 70(10), 1692–1699.
19. H. C. Wu, T. C. Chen, H. J. Tsai and C. S. Chen, Au nanoparticles deposited on magnetic carbon nanofibers as the ultrahigh sensitive substrate for surface-enhanced Raman scattering: Detections of rhodamine 6G and aromatic amino acids, Langmuir, 2018, 34(47), 14158–14168.
20. O. D. Prasiwi, T. E. Saraswati, M. Anwar and A. Masykur, Magnetic carbon nanofibers prepared with Ni and Ni/graphitic carbon nanoparticle catalysts for glycine detection using surface-enhanced Raman spectroscopy, ACS Appl. Nano Mater., 2021, 4(7), 6594–6608.
21. N. Sunil, U. Rajesh and B. Pullithadathil, Label-Free SERS Salivary Biosensor Based on Ni@Ag Core–Shell Nanoparticles Anchored on Carbon Nanofibers for Prediagnosis of Lung Cancer, ACS Appl. Nano Mater., 2023, 6(13), 11334–11350.
22. Keerthi G. Nair, Vishnuraj Ramakrishnan, Rajesh Unnathpadi, Karthikeyan K. Karuppanan and Biji Pullithadathil, Unraveling hydrogen adsorption kinetics of bimetallic Au–Pt nanoisland-functionalized carbon nanofibers for room-temperature gas sensor applications, J. Phys. Chem. C, 2020, 124(13), 7144–7155.
23. S. Shang, A. Kunwar, Y. Wang, X. Qi, H. Ma and Y. Wang, Synthesis of Cu@Ag core–shell nanoparticles for characterization of thermal stability and electric resistivity, Appl. Phys. A: Mater. Sci. Process., 2018, 124, 1–8.
24. A. G. El-Deen, N. A. Barakat, K. A. Khalil and H. Y. Kim, Hollow carbon nanofibers as an effective electrode for brackish water desalination using the capacitive deionization process, New J. Chem., 2014, 38(1), 198–205.
25. M. Sadhucharan, P. Sanpui, S. Sankar Ghosh, A. Chattopadhyay and A. Paul, Synthesis, characterization and enhanced bactericidal action of a chitosan supported core–shell copper–silver nanoparticle composite, RSC Adv., 2015, 5(16), 12268–12276.
26. Z. Jun, D. Zhang and J. Zhao, Fabrication of Cu–Ag core–shell bimetallic superfine powders by eco-friendly reagents and structures characterization, J. Solid State Chem., 2011, 184(9), 2339–2344.
27. S. Petrović, B. Salatić, D. Milovanović, V. Lazović, L. Živković, M. Trtica and B. Jelenković, Agglomeration in core-shell structure of CuAg nanoparticles synthesized by the laser ablation of Cu target in aqueous solutions, J. Opt., 2015, 17(2), 025402.
28. A. Sakthisabarimoorthi, M. Jose, S. A. Martin Britto Dhas and S. Jerome Das, Fabrication of Cu@ Ag core–shell nanoparticles for nonlinear optical applications, J. Mater. Sci.: Mater. Electron., 2017, 28, 4545–4552.
29. N. R. Barveen, T. J. Wang and Y. H. Chang, In-situ deposition of silver nanoparticles on silver nanoflowers for ultrasensitive and simultaneous SERS detection of organic pollutants, Microchem. J., 2020, 159, 105520.
30. E. A. Kumar, N. R. Barveen, T. J. Wang, T. Kokulnathan and Y. H. Chang, Development of SERS platform based on ZnO multipods decorated with Ag nanospheres for detection of 4-nitrophenol and rhodamine 6G in real samples, Microchem. J., 2021, 170, 106660.
31. S. Shengyan, A. Kunwar, Y. Wang, X. Qi, H. Ma and Y. Wang, Synthesis of Cu@ Ag core–shell nanoparticles for characterization of thermal stability and electric resistivity, Appl. Phys. A, 2018, 124, 1–8.
32. L. Retterstol, T. Lyberg, T. Aspelin and K. Berg, A twin study of nitric oxide levels measured by serum nitrite/nitrate, Twin Res. Hum. Genet., 2006, 9(2), 210–214.
33. R. Weller, Nitric oxide—a newly discovered chemical transmitter in human skin, Br. J. Dermatol., 1997, 137(5), 665–672.
34. H. Yang, Y. Xiang, X. Guo, Y. Wu, Y. Wen and H. Yang, Diazo-reaction-based SERS substrates for detection of nitrite in saliva, Sens. Actuators, B, 2018, 271, 118–121.
35. N. Sunil, R. Unnathpadi and B. Pullithadathil, Ag Nanoislands Functionalized Hollow Carbon Nano Fibers as Non-Invasive, Label-Free SERS Salivary Biosensor Platform for Salivary Nitrite Detection for Pre-Diagnosis of Oral Cancer, Analyst, 2024, 149(17), 4443–4453.
36. X. M. Qu, Z. F. Wu, B. X. Pang, L. Y. Jin, L. Z. Qin and S. L. Wang, From nitrate to nitric oxide: the role of salivary glands and oral bacteria, J. Dent. Res., 2016, 95(13), 1452–1456.
37. V. Shende, A. T. Biviji and N. Akarte, Estimation and correlative study of salivary nitrate and nitrite in tobacco related oral squamous carcinoma and submucous fibrosis, J. Oral Maxillofac. Pathol., 2013, 17(3), 381.
38. S. Navami, R. Unnathpadi and B. Pullithadathil, Silver Anchored α-MnO₂ Nanorods Based SERS Substrates for Salivary Thiocyanate Detection and Application in Oral Cancer Diagnosis, J. Biomed. Photonics Eng., 2023, 9(3), 030311.
39. K. Tsuge, M. Kataoka and Y. Seto, Cyanide and thiocyanate levels in blood and saliva of healthy adult volunteers, J. Health Sci., 2000, 46(5), 343–350.
40. K. Timo and G. D. Bothun, In situ SERS detection of dissolved nitrate on hydrated gold substrates, Nanoscale Adv., 2021, 3(14), 4098–4105.
41. G. Shashikanth, C. Fan, M. Lin and Z. Hu, Quantitative detection of nitrate in water and wastewater by surface-enhanced Raman spectroscopy, Environ. Monit. Assess., 2013, 185, 5673–5681.
42. W. Jingjing, Md Mehedi Hassan, W. Ahmad, T. Jiao, Y. Xu, H. Li, Q. Ouyang, Z. Guo and Q. Chen, A highly structured hollow ZnO@Ag nanosphere SERS substrate for sensing traces of nitrate and nitrite species in pickled food, Sens. Actuators, B, 2019, 285, 302–309.
43. Y. Hanru, Y. Xiang, X. Guo, Y. Wu, Y. Wen and H. Yang, Diazo-reaction-based SERS substrates for detection of nitrite in saliva, Sens. Actuators, B, 2018, 271, 118–121.
44. A. Ianoul, T. Coleman and S. A. Asher, UV resonance Raman spectroscopic detection of nitrate and nitrite in wastewater treatment processes, Anal. Chem., 2002, 74(6), 1458–1461.
45. Y. Feng, R. Mo, L. Wang, C. Zhou, P. Hong and C. Li, Surface enhanced Raman spectroscopy detection of sodium thiocyanate in milk based on the aggregation of Ag nanoparticles, Sensors, 2019, 19(6), 1363.
46. A. Nizamuddin, F. Arith, J. Rong, M. Zaimi, A. S. Rahimi and S. Saat, Investigation of copper (I) thiocyanate (CuSCN) as a hole transporting layer for perovskite solar cells application, J. Adv. Res. Fluid Mech. Therm. Sci., 2020, 78(2), 153–159.
47. J. M. Connolly, K. Davies, A. Kazakeviciute, A. M. Wheatley, P. Dockery, I. Keogh and M. Olivo, Non-invasive and label-free detection of oral squamous cell carcinoma using saliva surface-enhanced Raman spectroscopy and multivariate analysis, Nanomed.: Nanotechnol. Biol. Med., 2016, 12(6), 1593–1601.

---

Footnote

† Electronic supplementary information (ESI) available: The SEM images of CNFs, Cu/CNFs, and Cu@Ag/CNF, TEM analysis of Cu@Ag/CNFs at different ratios of Cu : Ag = 1 : 1, 1 : 3 and 1 : 5 and the inset images show the core shell structure at different molar ratios. Video of the microfluidic chip based system for saliva sample analysis. See DOI: https://doi.org/10.1039/d4tb02766c

---