Method for blood cfDNA analysis based on circularly polarized surface-enhanced displacement raman differential spectroscopy

Abstract

This invention discloses a blood cfDNA analysis method based on circularly polarized surface-enhanced displacement Raman differential spectroscopy, relating to the field of blood cfDNA detection. The method includes: for each cfDNA sample from a non-recurrent nasopharyngeal carcinoma patient and a cfDNA sample from a recurrent nasopharyngeal carcinoma patient, using two slightly offset wavelengths as the excitation source for a circularly polarized Raman probe, obtaining LHCP Raman differential spectral images and RHCP Raman differential spectral images through left-handed and right-handed circularly polarized filters, decomposing and reconstructing them into background-free LHCP and RHCP spectral images, and then fusing the background-free LHCP and RHCP spectral images to obtain a fused image; performing principal component analysis on the fused image to reduce the dimensionality to retain the significant principal components with the largest variance associated with the recurrent and non-recurrent groups. This invention is the first to combine a circularly polarized Raman system with surface-enhanced displacement Raman differential spectroscopy, and maximizes spectral discrimination capability through fused spectral data, better identifying subtle molecular changes in cfDNA.

Description

Technical Field

[0001] This invention relates to the field of blood cfDNA detection, and in particular to a blood cfDNA analysis method based on circularly polarized surface-enhanced displacement-excited Raman differential spectroscopy. Background Technology

[0002] Nasopharyngeal carcinoma (NPC) is a common head and neck malignancy, with a significantly higher incidence in Southeast Asia and southern China. Radiotherapy combined with chemotherapy is a primary treatment method, achieving high local control rates for NPC patients. However, recurrence remains a serious concern. Statistics show that approximately 25% of patients with locally advanced NPC experience recurrence within 2-3 years after standard treatment, and the 5-year survival rate drops sharply after recurrence. Therefore, early and accurate detection of NPC recurrence has become a crucial clinical need, helping physicians develop more individualized treatment plans and improve patient prognosis.

[0003] Currently, predicting nasopharyngeal carcinoma recurrence before treatment mainly relies on clinical staging systems and blood biomarker testing. The tumor-lymph node-metastasis (TNM) staging system based on traditional CT / MR imaging initially assesses recurrence risk through tumor size and lymph node metastasis. However, this system relies solely on anatomical features and cannot capture tumor molecular heterogeneity, thus its predictive accuracy is insufficient. This problem manifests in the significant differences in prognosis among patients with the same clinical TNM stage who receive the same treatment. While traditional blood biomarkers (such as EBV VCA-IgA antibodies and EBV DNA) can be obtained non-invasively, their sensitivity is insufficient to serve as reliable independent indicators, making it difficult to accurately identify high-risk patients and guide individualized treatment.

[0004] Circulating cell-free DNA (cfDNA) detection, as an emerging non-invasive molecular diagnostic tool, offers new prospects for the early detection and prognosis prediction of nasopharyngeal carcinoma. When tumor cells undergo apoptosis or necrosis, they release cfDNA into the bloodstream. cfDNA carries information such as tumor-specific gene mutations and methylation modifications, and theoretically could serve as an ideal biomarker for recurrence monitoring.

[0005] However, current cfDNA detection mainly relies on polymerase chain reaction (PCR) and next-generation sequencing (NGS) technologies. Traditional PCR lacks stability and accuracy, while NGS has limitations such as high cost, long processing time, and stringent operator skill requirements, making it unsuitable for large-scale nasopharyngeal carcinoma detection. Therefore, developing a simple, efficient, and accurate cfDNA detection method for predicting recurrence before nasopharyngeal carcinoma treatment has significant clinical value. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a blood cfDNA analysis method based on circularly polarized surface-enhanced displacement Raman differential spectroscopy. This method combines circularly polarized Raman technology with surface-enhanced displacement Raman differential spectroscopy for the first time, enabling the acquisition of high-quality, fluorescence-free circularly polarized surface-enhanced displacement Raman differential spectra of blood cfDNA. By fusing spectral data with LHCP+RHCP, the spectral discrimination capability is maximized, allowing for better identification of subtle molecular changes in cfDNA. This method can be applied to analyze the differential biochemical fingerprint characteristics of blood cfDNA in patients with recurrent and non-recurrent nasopharyngeal carcinoma before treatment.

[0007] This invention provides a method for analyzing blood cfDNA based on circularly polarized surface-enhanced displacement-excited Raman differential spectroscopy, comprising: Pre-treatment plasma samples from non-recurrent nasopharyngeal carcinoma patients and pre-treatment plasma samples from recurrent nasopharyngeal carcinoma patients were obtained, and cfDNA samples from non-recurrent nasopharyngeal carcinoma patients and cfDNA samples from recurrent nasopharyngeal carcinoma patients were prepared. For each non-recurrent nasopharyngeal carcinoma (NPC) patient's cfDNA sample and each recurrent NPC patient's cfDNA sample, two slightly offset wavelengths were used as the excitation source for a circularly polarized Raman probe. Two Raman differential spectral images at different wavelengths were acquired at the same sample location using the circularly polarized Raman probe. The excitation source was passed through a left-handed circularly polarized filter to obtain an LHCP Raman differential spectral image, and the excitation source was passed through a right-handed circularly polarized filter to obtain an RHCP Raman differential spectral image. The corresponding differential spectra were obtained through iterative calculation using a differential spectral decomposition method, and then reconstructed into background-free LHCP and RHCP spectral images. The background-free LHCP and RHCP spectral images were then fused to obtain a fused image. Principal component analysis is performed on the fused image to reduce its dimensionality to significant principal components, which retain the maximum variance associated with the relapse and non-relapse groups.

[0008] Furthermore, the method also includes integrating significant principal components using linear discriminant analysis and constructing a linear decision boundary based on the distance from the sample to the class centroid to obtain the LDA posterior probability distribution map of recurrent and non-recurrent nasopharyngeal carcinoma categories.

[0009] Furthermore, the method also includes constructing a classification model for predicting recurrence of nasopharyngeal carcinoma before treatment using machine learning methods, and training the model using the fusion spectra of non-recurrent nasopharyngeal carcinoma patients and the fusion spectra of recurrent nasopharyngeal carcinoma patients as training data. Pretreatment plasma samples from the target patient were obtained and converted into cfDNA samples. A fused image of the target patient was acquired using a circularly polarized Raman probe combined with displacement-excited Raman differential spectroscopy. The fused image was then input into a classification model to obtain the classification result.

[0010] Furthermore, the LHCP spectral image and the RHCP spectral image are fused to obtain a fused image. Specifically, the LHCP spectral image and the RHCP spectral image are normalized separately and then horizontally stitched together to obtain a fused image.

[0011] Furthermore, the illumination arm of the circularly polarized Raman probe includes a collimator, a bandpass filter, a left-handed polarizer, a right-handed polarizer, and a focusing lens; the collection arm includes a collection lens, a long-pass filter, and a refocusing lens. The collected light is transmitted to a spectrometer for spectral analysis via an optical fiber bundle. The cfDNA spectral acquisition range is 410-1610 cm⁻¹. -1 .

[0012] Furthermore, the differential spectral decomposition method simultaneously utilizes two original spectra and their inherent correlation to distinguish between Raman signals and interference signals.

[0013] Furthermore, plasma cfDNA was extracted from pre-treatment plasma samples of non-recurrent nasopharyngeal carcinoma patients and pre-treatment plasma samples of recurrent nasopharyngeal carcinoma patients, and dissolved in nuclease-free water to obtain cfDNA solutions. The silver nanoparticle mixture was mixed with the cfDNA solution, then dropped onto a substrate and dried to obtain cfDNA samples of non-recurrent nasopharyngeal carcinoma patients and cfDNA samples of recurrent nasopharyngeal carcinoma patients.

[0014] The technical solutions provided in the embodiments of the present invention have at least the following technical effects: 1. This study is the first to explore the integration of circular polarization technology with surface-enhanced displacement-excited Raman differential spectroscopy (SE-SERDS) to obtain high-quality, fluorescence-free circularly polarized surface-enhanced displacement-excited Raman differential spectra. Compared with traditional polynomial fitting methods, it achieves better autofluorescence subtraction and can extract high-quality cfDNA Raman signals, including more subtle peaks related to the prognosis of nasopharyngeal carcinoma. It has great potential for prognostic prediction and guiding personalized treatment in precision oncology.

[0015] 2. Circular polarization excitation improves the discrimination ability by capturing the chiral structural changes of cfDNA. Compared with non-polarization excitation, it produces more significant spectral differences between the relapse and non-relapse groups. More importantly, through signal complementarity integration, the LHCP+RHCP fusion spectrum is superior to the RHCP or LHCP modes alone (accuracy of 93.9% vs. 88.9% and 81.7% under SE-SERDS combined with LDA conditions).

[0016] 3. For the first time, circular polarization technology (including LHCP+RHCP spectral fusion) was integrated with SE-SERDS, and its synergistic advantages significantly improved the accuracy of prognostic prediction for nasopharyngeal carcinoma. This innovative method has great potential as a novel and promising tool for predicting prognostic outcomes before cancer treatment.

[0017] 4. The synergistic integration of LHCP+RHCP with SE-SERDS and SVM models achieved the best prediction accuracy (95.6%). When applied to blood cfDNA analysis, it has the advantages of rapid detection, low cost and high accuracy. It is suitable as a new biosensing tool to predict the recurrence type of nasopharyngeal carcinoma patients, accurately identify high-risk patients and guide individualized treatment.

[0018] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Figure 1 This is a schematic diagram of the system framework upon which the method of the embodiments of the present invention is based; Figure 2 This is a general flowchart of the method according to an embodiment of the present invention; Figure 3 This is a comparison of normalized PF-SERS and SE-SERDS spectra of blood cfDNA from nasopharyngeal carcinoma patients under laser excitation with different polarization states in this embodiment of the invention; Figure 4 Box plots of the intensity of ten significant SE-SERDS peaks in the blood cfDNA of patients with recurrent and non-recurrent nasopharyngeal carcinoma under different polarization laser excitation in this embodiment of the invention; Figure 5 These are principal component loading diagrams of SE-SERDS spectra under NP, LHCP, RHCP, and LHCP+RHCP fusion conditions, respectively, in embodiments of the present invention. Figure 6 This is a comparative analysis of the detection results of PF-SERS and SE-SERDS under non-polarized, left-handed circularly polarized, right-handed circularly polarized, and fused spectral conditions in the embodiments of the present invention. Figure 7 This is a bar chart showing the average principal component score of SE-SERDSD under non-polarized, left-handed circularly polarized, right-handed circularly polarized, and fused spectral conditions in an embodiment of the present invention. Figure 8The image shows the detection results based on the LHCP+RHCP+NP fused SE-SERDS spectrum in an embodiment of the present invention. Figure 9 This is a flowchart illustrating the preparation of a blood cfDNA-silver nanoparticle mixture and the use of circularly polarized SE-SERDS spectroscopy for predicting recurrence before nasopharyngeal carcinoma treatment in an embodiment of the present invention. Figure 10 The confusion matrices of the LR, KNN, and SVM models in the embodiments of the present invention, as well as the ROC curves of the three spectral datasets (LHCP, RHCP, and LHCP+RHCP) under each model; Figure 11 The statistical performance metrics of four machine learning models (LDA, LR, KNN, SVM) in this embodiment of the invention on three spectral datasets are shown. Figure 12 This invention demonstrates the prognostic predictive performance of PF-SERS and SE-SERDS spectra under different polarization modes for pretreatment differential diagnosis of nasopharyngeal carcinoma (NPC) recurrent group (Rec, n=90) and non-recurrent group (Nonrec, n=90). Detailed Implementation

[0021] This invention provides a blood cfDNA analysis method based on circularly polarized surface-enhanced displacement Raman (SVM) spectroscopy. For the first time, it combines a circularly polarized Raman system with SVM spectroscopy to acquire high-quality, fluorescence-free SVM spectroscopy data of blood cfDNA. By fusing spectral data, it maximizes spectral discrimination capabilities, better identifying subtle molecular changes in cfDNA. This method can be applied to analyze the differential biochemical fingerprint characteristics of blood cfDNA in patients with recurrent and non-recurrent nasopharyngeal carcinoma before treatment.

[0022] The overall concept of the technical solutions in the embodiments of the present invention is as follows: Raman spectroscopy is a novel analytical technique with advantages such as being non-invasive, relatively inexpensive, and requiring minimal operator training. However, traditional Raman spectroscopy (CRS) has inherent limitations: extremely small molecular scattering cross sections and strong autofluorescence interference in biological samples result in weak Raman signals that are often masked by fluorescence, severely hindering its practical application in medical testing.

[0023] To address the issue of weak signals, surface-enhanced Raman scattering (SERS) has emerged as a reliable alternative. When target molecules approach silver nanoparticles, SERS can enhance the Raman signal by approximately 10⁻⁶. 14 The SERS combined with conventional polynomial fitting (PF) has been used for surface-enhanced Raman spectroscopy detection of plasma and has achieved good results.

[0024] However, circulating cell-free DNA and human blood cells are on completely different orders of magnitude in size; the former is at the molecular level in length, while the latter is at the cellular level in diameter. For example, a typical 7μm red blood cell has a diameter equivalent to a DNA chain length of approximately 20,000 base pairs (bp). Although surface-enhanced Raman spectroscopy (PF-SERS) based on traditional polynomial fitting (PF) correction can also detect circulating cell-free DNA, its detection accuracy is insufficient for accurately predicting recurrence before nasopharyngeal carcinoma treatment.

[0025] This application explores the combination of circular polarization technique (LHCP+RHCP fusion spectroscopy) and surface-enhanced displacement-excited Raman differential spectroscopy (SE-SERDS) for cfDNA detection, a field with no previously reported research. Therefore, this application aims to develop a blood cfDNA analysis method based on circular polarization surface-enhanced displacement-excited Raman differential spectroscopy and to validate its sensitivity, specificity, and accuracy in predicting recurrence before nasopharyngeal carcinoma treatment.

[0026] Before introducing specific embodiments, the system framework corresponding to the method of this application is first described. The system is roughly divided into two parts: a circularly polarized Raman subsystem and a differential control subsystem. The circularly polarized Raman subsystem includes a circularly polarized Raman probe and a spectrometer; the differential control subsystem includes a dual-wavelength excitation source and a control and forwarding module; the dual-wavelength excitation source is connected to the circularly polarized Raman probe; the control and forwarding module is connected to the dual-wavelength excitation source, the spectrometer, and the main control unit; the control and forwarding module controls the dual-wavelength excitation source to generate two wavelengths of excitation light at specified time intervals according to the detection instructions of the main control unit, so that the circularly polarized Raman subsystem transmits circularly polarized laser light to the detection sample, and obtains the corresponding circularly polarized Raman spectrum from the spectrometer, and then forwards it to the main control unit.

[0027] like Figure 1 As shown, the circularly polarized Raman probe includes an illumination arm, a collection arm, a left-handed circularly polarized filter, and a right-handed circularly polarized filter. The illumination arm includes a collimator, a bandpass filter (BP filter), a polarizer, and a focusing lens, used to transmit laser light of different circularly polarized states (laser spot size 3 mm) to the sample. The collection arm includes a collection lens, a long-pass filter (LP filter), and a refocusing lens, which transmits the collected light to the spectrometer for spectral analysis via an optical fiber bundle (fiber core diameter 100 μm, numerical aperture NA = 0.22).

[0028] The excitation source is a dual-wavelength diode laser with wavelengths of 784.5 nm and 785.2 nm, which are excited at specified intervals.

[0029] This application innovatively integrates circularly polarized Raman spectroscopy with surface-enhanced displacement-excited Raman differential spectroscopy for the first time, achieving high-quality, fluorescence-free circularly polarized SE-SERDS spectroscopy acquisition of cfDNA. Before measurement, the system was calibrated using a neon-argon (Ne-Ar) standard lamp (Princeton Instruments) and a silicon wafer. The cfDNA spectral acquisition range is 410–1610 cm⁻¹. -1 Spectral resolution of 8cm -1 The integration time is 5 seconds. Example 1

[0030] This embodiment provides a blood cfDNA analysis method based on circularly polarized surface-enhanced displacement-excited Raman differential spectroscopy, such as... Figure 2 As shown, it includes: S1. cfDNA Sample Preparation Process: Pre-treatment plasma samples from patients with non-recurrent nasopharyngeal carcinoma and patients with recurrent nasopharyngeal carcinoma were obtained to prepare cfDNA samples from both patients. The preparation process was as follows: cfDNA was extracted from both pre-treatment plasma samples and pre-treatment plasma samples from patients with non-recurrent and recurrent nasopharyngeal carcinoma, and dissolved in nuclease-free water to obtain cfDNA solutions. A mixture of silver nanoparticles was mixed with the cfDNA solutions, then dropped onto a substrate and dried to obtain cfDNA samples from both patients.

[0031] In one specific embodiment, 5 mL of peripheral blood was collected from each patient upon diagnosis. Whole blood (EDTA anticoagulated) was centrifuged at 1600 × g for 10 minutes at 4°C to separate plasma. The plasma was then centrifuged at 16000 × g for 10 minutes to remove cell debris. Enzyme-free centrifuge tubes were used throughout the process, and the processed plasma was stored at -80°C for analysis.

[0032] Plasma cfDNA was extracted using a magnetic bead method and a cfDNA extraction kit on a Pre-NAT fully automated workstation, ensuring standardized and high-throughput processing. Finally, the isolated cfDNA was dissolved in 60 μL of nuclease-free water and stored at -20°C to prevent degradation for subsequent spectroscopic analysis.

[0033] Silver nanoparticles (AgNPs) were used as the SERS substrate. The synthesis process was as follows: 4.5 mL of 0.1 mol / L sodium hydroxide solution was mixed with 5 mL of 0.06 mol / L hydroxylamine hydrochloride, and then quickly poured into 90 mL of 0.0011 mol / L silver nitrate solution. The mixture was stirred at room temperature until the solution turned milky gray. Subsequently, the synthesized silver nanoparticle colloid was centrifuged at 10,000 rpm for 10 minutes, and the supernatant was discarded to obtain a concentrated colloid. Characterization results showed that the silver nanoparticles exhibited an absorption peak at 417 nm; the zeta potential was -16.22 mV; dynamic light scattering and transmission electron microscopy (TEM) images showed that the particle size followed a normal distribution, with an average particle size of approximately 50 nm.

[0034] Before measurement, 5 μL of silver nanoparticles were mixed with 5 μL of cfDNA solution and stored at 4°C; then the mixture was dropped onto a grooved pure aluminum plate, dried, and then subjected to SERS detection.

[0035] S2. Circularly Polarized Raman Differential Spectroscopy Fusion Image Acquisition Process: For each non-recurrent nasopharyngeal carcinoma (NPCC) patient's cfDNA sample and a recurrent NPCC patient's cfDNA sample, two slightly offset wavelengths are used as the excitation source for the circularly polarized Raman probe. Two Raman differential spectral images at different wavelengths are continuously acquired at the same sample location using the circularly polarized Raman probe. The excitation source is passed through a left-handed circularly polarized filter to obtain the LHCP Raman differential spectral image, and the excitation source is passed through a right-handed circularly polarized filter to obtain the RHCP Raman differential spectral image. The corresponding differential spectra are obtained by iterative calculation using the differential spectral decomposition method, and then the differential spectra are reconstructed into background-free LHCP and RHCP spectral images. The background-free LHCP and RHCP spectral images are fused to obtain the fused image.

[0036] S3. Principal component analysis process: Principal component analysis is performed on the fused image to reduce the dimensionality of the complex spectral data into significant principal components. The significant principal components retain the maximum variance associated with the relapse group and the non-relapse group.

[0037] This application innovatively applies the circularly polarized Raman subsystem and SE-SERDS to analyze the differential biochemical fingerprint characteristics of blood cfDNA in patients with recurrent and non-recurrent nasopharyngeal carcinoma before treatment. Circularly polarized excitation identified more significant differential peaks, reflecting the potential of this technology in more accurately predicting nasopharyngeal carcinoma recurrence.

[0038] The differential spectral decomposition method utilizes two original spectra and their intrinsic correlation to distinguish Raman signals from interference signals. The original spectra generated by laser excitation are designated as Spectrum 1 (target curve) and Spectrum 2. Three curves are generated by superposition to fit Spectrum 1. The differential spectrum is obtained through iterative calculation, and then reconstructed to obtain the background-free SE-SERDS spectrum.

[0039] Figure 3 The normalized PF-SERS and SE-SERDS spectra of blood cfDNA from patients with recurrent nasopharyngeal carcinoma were compared under NP (unpolarized), LHCP (left-handed circularly polarized), and RHCP (right-handed circularly polarized) laser excitation. Figure 3 (A and B in the text). It is worth noting that SE-SERDS is at 1074cm. -1 and 1147cm -1 Additional changes have also occurred ( Figure 3 (marked by the red arrow in B). Figure 3 The CH images present the mean PF-SERS and SE-SERDS spectra of blood cfDNA from 90 patients with recurrent and 90 without recurrent nasopharyngeal carcinoma under three excitation conditions (NP, LHCP, and RHCP). The shaded areas represent the standard deviation of spectral intensity. Comparative analysis of the spectral data for each technique (PF-SERS or SE-SERDS alone) revealed that the spectral profiles and positions of characteristic Raman peaks were consistent between the two groups. The characteristic peaks involved included those at 430, 479, 540, 673, 780, 923, 1030, 1179, 1300, 1404, and 1475 cm⁻¹. -1 (Among them, 1074 and 1147cm) -1 (Appears only in SE-SERDS). However, some peaks appear between the two groups (e.g., 430, 780, 923, 1030, 1074, 1147, 1179, 1300, 1404, and 1475 cm⁻¹). -1 The relative intensities of the two groups differ significantly, a phenomenon observed in... Figure 3 The difference spectra (recurrence group minus non-recurrence group) at the bottom of each CH subplot are visually displayed. Comparative analysis of the difference spectra of PF-SERS and SE-SERDS under the same polarization conditions revealed that the main peak positions of the two techniques are generally consistent, but SE-SERDS has a significant advantage: it resolves additional subtle difference peaks (such as 1074 and 1147 cm⁻¹) that PF-SERS cannot detect. -1 Furthermore, under circular polarization conditions, its difference peaks are more significant than those of PF-SERS. These observations indicate that SE-SERDS has a superior ability to distinguish between groups and prognostic spectral differences compared to PF-SERS.

[0040] To further evaluate the impact of circular polarization on the spectral discrimination capability of SE-SERDS (SE-SERDS has been proven to be a superior technique), a systematic quantitative comparison of SE-SERDS spectra was conducted under NP, LHCP, and RHCP conditions. Figure 4Box plots of ten significant SE-SERDS peak intensities were presented, comparing the differences between recurrent and non-recurrent nasopharyngeal carcinoma groups under three polarization excitation conditions. Under non-polarization excitation, significant differences in peak intensity were observed between the two groups: 430, 780, 1147, and 1179 cm⁻¹. -1 The peak values ​​were lower in the relapse group than in the non-relapse group, with values ​​of 923, 1030, 1074, 1300, and 1404 cm⁻¹. -1 The peaks were higher in the relapsed group than in the non-relapsed group. Similar intensity differences were also detected in these peaks under circularly polarized (LHCP and RHCP) excitation, with one significant exception: under RHCP excitation, the peaks at 780, 1030, and 1475 cm⁻¹ were higher. -1 The peaks exhibit completely opposite intensity trends between the two groups. Under all excitation conditions, a consistent pattern emerges: the SE-SERDS spectra of circularly polarized excitation contain more peaks with statistically significant differences (p < 0.05) between groups, including peaks at 923, 1030, 1147, 1179, 1404, and 1475 cm⁻¹. -1 (Unpolarized excitation only 430, 780, 1074 cm⁻¹) -1 (Significant). Additionally, 923, 1300, and 1475 cm... -1 The p-values ​​of the peaks at LHCP were lower than those at RHCP, while those at 780, 1030, and 1179 cm⁻¹ were lower. -1 The p-value of the peak at RHCP is lower than that at LHCP, and the lower the p-value, the more significant the difference between groups.

[0041] In this embodiment, 780cm -1 and 1030cm -1 The peaks (preliminarily assigned to C and A bases) exhibited different vibrational intensities between the recurrent and non-recurrent nasopharyngeal carcinoma groups, suggesting abnormal metabolism of these two DNA bases in the blood of recurrent patients. Consistently, significant changes in these two peaks have also been observed in existing SERS-based studies of DNA methylation and mutation detection. This consistency stems from the fact that methylation and mutation disrupt the native structure of DNA, ultimately leading to alterations in the structure and composition of C and A bases.

[0042] Furthermore, compared to the relatively broad and indistinct peaks in PF-SERS, SE-SERDS spectra show sharper and more abundant peaks (such as at 1074 cm⁻¹). -1 and 1147cm -1The results (preliminarily attributed to T bases and PO2- stretching vibrations of DNA) indicate that SE-SERDS effectively mitigates background interference and increases the number of specific Raman signals for cfDNA. This improved spectral resolution allows for better identification of subtle molecular changes in cfDNA, such as nucleotide base pairing and DNA conformational alterations (e.g., B-DNA versus other forms), which are crucial for differentiating between recurrent and non-recurrent nasopharyngeal carcinoma patients.

[0043] More importantly, as the physiological state changes during cancer progression, the supercoiled structure of DNA alters, affecting radiosensitivity and driving local conformational shifts in some DNA molecules (such as from right-handed B-DNA to left-handed Z-DNA), thereby influencing polarization-dependent Raman scattering. Notably, this change in chiral characteristics may be indistinguishable under non-polarized excitation, while circularly polarized lasers generate a chiral supercoiled electromagnetic field that exhibits enhanced asymmetric interactions with chiral DNA molecules, making them sensitive to changes in molecular chirality. Furthermore, given that LHCP and RHCP each capture partial chiral signal information of the DNA structure, the inventors considered that fusing spectral data from the two polarization modes might yield more comprehensive information on cfDNA conformational changes. This embodiment innovatively fuses LHCP and RHCP spectral images; specifically, after normalizing the LHCP and RHCP spectral images respectively, they are horizontally stitched together to obtain a fused image. Other methods can also be used to fuse LHCP and RHCP spectral images (such as fusing the two normalized images into one image through mathematical operations), but the fusion method in this embodiment achieves the best principal component analysis results, which will be analyzed and explained in detail below.

[0044] The normalized PF-SERS and SE-SERDS spectra of cfDNA samples under NP, LHCP and RHCP photoexcitation were used to obtain six corresponding datasets. The LHCP and RHCP spectra were then fused using the method in this embodiment to obtain the PF-SERS fused dataset and the SE-SERDS fused dataset.

[0045] PCA-LDA is a simple and effective statistical method for spectral analysis in cancer identification. With the aid of an independent samples t-test (significance level p < 0.05), PCA reduces the dimensionality of complex spectral data to key principal components (PCs), which retain the largest variance associated with the relapse and non-relapse groups. In all eight datasets, each single-polarization dataset (NP, LHCP, RHCP) had three principal components showing significant differences between the two prognostic groups (p < 0.05), while the LHCP+RHCP fusion dataset produced five significant principal components due to enhanced spectral complementarity. Figure 5The correlation between these principal components and their spectral contributions was revealed. For the PF-SERS dataset, the significant principal components under unpolarized conditions are (PC1, PC3, PC5), under LHCP they are (PC1, PC2, PC3), under RHCP they are (PC2, PC3, PC4), and under the LHCP+RHCP fusion spectrum they are (PC1, PC2, PC3, PC4, PC5). For the SE-SERDS dataset, the significant principal components under unpolarized conditions are (PC3, PC5, PC7), under LHCP they are (PC1, PC2, PC3), under RHCP they are (PC2, PC3, PC4), and under the LHCP+RHCP fusion spectrum they are (PC1, PC2, PC3, PC12, PC13).

[0046] Focusing on SE-SERDS, the principal components from single LHCP and RHCP spectra cover Figure 4 All diagnostically relevant peaks (p < 0.05) were observed, including those at 430, 780, 923, 1030, 1074, 1147, 1179, 1300, 1404, and 1475 cm⁻¹. -1 In contrast, the non-polarization-derived principal components captured fewer of these discriminative features. Furthermore, the principal components of the LHCP+RHCP fused spectrum integrated all the discriminative features from the two single-polarization (LHCP, RHCP) datasets, demonstrating the complementarity of dual-polarization fusion. The variance explained by each principal component is as follows: non-polarization (PC3: 4.58%, PC5: 3.39%, PC7: 2.97%); LHCP (PC1: 9.76%, PC2: 5.76%, PC3: 5.06%); RHCP (PC2: 6.57%, PC3: 4.64%, PC4: 4.15%); LHCP+RHCP (PC1: 7.06%, PC2: 4.83%, PC3: 3.68%, PC12: 1.75%, PC13: 1.68%).

[0047] Figure 6 The scatter plots of significant principal components of PF-SERS (AC) and SE-SERDS (EG) spectra under single polarization (NP, LHCP, RHCP) conditions are shown. Figure 6Figures D and H in the table show the scatter plots of the LHCP+RHCP fusion spectra (producing five significant principal components) based on the three most significant principal components. It can be seen that in the PF-SERS and SE-SERDS scatter plots, the inter-group overlap is significant under non-polarized excitation; in contrast, the overlap is significantly reduced under RHCP excitation, and the relapsed and non-relapsed groups show a clear separation trend. Notably, the LHCP+RHCP fusion spectra achieve minimal inter-group overlap under all conditions, thus enabling more effective differentiation between the two groups. For SE-SERDS, Figure 7 These differences were quantified by bar charts of average principal component scores. The p-values ​​of significant principal components (all < 0.05, expressed in the order of Ex) showed a decreasing trend under non-polarized, LHCP, RHCP, and LHCP+RHCP conditions. In particular, under LHCP+RHCP conditions, the p-value of SE-SERDS was as low as E-19, which together confirmed its excellent prognostic discrimination ability.

[0048] Besides the approach in this embodiment of fusing LHCP and RHCP spectra before performing principal component analysis, it's also possible to first reduce the dimensionality of LHCP and RHCP spectra separately using PCA or PLS, and then fuse the resulting principal components. PCA is an unsupervised linear dimensionality reduction method. Its goal is to find the direction of maximum variance (principal component) in the data without relying on any labels, projecting the original high-dimensional data into a low-dimensional space. PLS is a supervised linear dimensionality reduction and regression method. Its goal is to maximize the covariance between X and Y (labels) while reducing dimensionality, i.e., finding the potential components that can both summarize the information of X and predict Y. The results obtained by dimensionality reduction followed by fusion are different from those obtained by fusion followed by dimensionality reduction, and this approach can serve as a supplement to other implementation methods. Verification has shown that fusing first and then performing principal component analysis is more valuable for predicting the recurrence type of nasopharyngeal carcinoma.

[0049] It is worth noting that supplementary validation showed that adding non-polarized spectroscopy (NP) to the LHCP+RHCP fusion of SE-SERDS did not further improve the discrimination performance. Figure 8The results show significant principal component scatter plots (A: PC1, PC2, PC3; B: PC1, PC4, PC6), LDA posterior probability distribution plots (C), and ROC curves (D) for nasopharyngeal carcinoma recurrent group (orange, Rec) and non-recurrent group (green, Nonrec) based on LHCP+RHCP+NP fusion SE-SERDS spectroscopy. The corresponding evaluation metrics are: sensitivity = 90.0%, specificity = 97.8%, accuracy = 93.9%, and AUC = 0.977. The LHCP+RHCP+NP fusion (accuracy: 93.9%, AUC: 0.977) is almost identical to the LHCP+RHCP SE-SERDS fusion alone (accuracy: 93.9%, AUC: 0.987). Figure 8 The D paper also presents the ROC curve (AUC=0.987) of the LHCP+RHCP fused SE-SERDS spectrum to show that the performance of the two is extremely close. This finding confirms that the improvement in accuracy specifically stems from the complementary chiral response between LHCP and RHCP, rather than a simple superposition of more spectral datasets. This application is the first to integrate circular polarization technology (including LHCP+RHCP spectral fusion) with SE-SERDS, leveraging their synergistic advantages to significantly improve the accuracy of nasopharyngeal carcinoma prognostic prediction. This innovative method has great potential as a novel and promising tool for pre-treatment prognostic prediction of cancer.

[0050] These findings confirm that optimizing spectral data (especially combining dual polarization fusion with SE-SERDS) can significantly improve the predictive performance of nasopharyngeal carcinoma recurrence.

[0051] To achieve quantitative identification, the method in this embodiment further includes: S4. Posterior probability statistical process: Linear discriminant analysis is used to integrate significant principal components, and a linear decision boundary is constructed based on the distance from the sample to the class centroid to obtain the LDA posterior probability distribution map of recurrent and non-recurrent nasopharyngeal carcinoma categories.

[0052] This embodiment uses 5-fold cross-validation LDA to integrate these significant principal components (3 single-polarization datasets and 5 LHCP+RHCP fusion datasets), and constructs a linear decision boundary based on the distance from the sample to the class centroid. Figure 6The IP shows the LDA posterior probability distribution of recurrent and non-recurrent nasopharyngeal carcinoma categories on the PF-SERS (IL) and SE-SERDS (MP) datasets under non-polarized, LHCP, RHCP, and LHCP+RHCP conditions. With a threshold of 0.5, the sensitivity, specificity, and accuracy of PF-SERS were as follows: non-polarized (71.1% (64 / 90), 62.2% (56 / 90), 66.7% (120 / 180)); LHCP (83.3% (75 / 90), 70.0% (63 / 90), 76.7% (138 / 180)); RHCP (90.0% (81 / 90), 80.0% (72 / 90), 85.0% (153 / 180)); LHCP+RHCP (91.1% (82 / 90), 84.4% (76 / 90), 87.8% (158 / 180)). The SE-SERDS spectra are as follows: unpolarized (72.2% (65 / 90), 70.0% (63 / 90), 71.1% (128 / 180)); LHCP (83.3% (75 / 90), 80.0% (72 / 90), 81.7% (147 / 180)); RHCP (90.0% (81 / 90), 87.8% (79 / 90), 88.9% (160 / 180)); LHCP+RHCP (91.1% (82 / 90), 96.7% (87 / 90), 93.9% (169 / 180)), with the LHCP+RHCP fusion SE-SERDS exhibiting the best spectral performance.

[0053] The corresponding ROC curve ( Figure 6 The QT (Queries and Tests) further quantified the classification performance of the models under all spectral conditions. For PF-SERS, the area under the curve (AUC) of RHCP was 0.902, which was better than LHCP (0.830) and unpolarized (0.759), while the AUC of the LHCP+RHCP fusion spectrum was even higher at 0.941. For SE-SERDS, the AUC of RHCP was 0.913, which exceeded LHCP (0.847) and unpolarized (0.784), and the LHCP+RHCP fusion spectrum showed the largest AUC (0.987) across all datasets. Overall, these results ( Figure 6 The statistical summary of the UW display further supports this view, showing three key trends: First, SE-SERDS consistently outperforms PF-SERS under all polarization conditions; second, RHCP outperforms LHCP and non-polarized excitation for both PF-SERS and SE-SERDS techniques; and third, the LHCP+RHCP fusion spectrum further enhances the discrimination performance.

[0054] This embodiment reveals key prognostic peaks (such as 1074 and 1147 cm⁻¹) that were previously masked by fluorescence interference. -1 The LHCP+RHCP fusion spectroscopy enriches the spectral signal with more detailed information to characterize nasopharyngeal carcinoma recurrence. Further improvements in the LHCP+RHCP fusion spectroscopy may stem from the complementary chiral response signals captured by the two polarizations (reflected in five significant principal components covering a wider range of spectral features), which together capture more comprehensive prognostic biomolecular information, achieving the highest predictive accuracy.

[0055] Preferably, to improve the accuracy of predicting nasopharyngeal carcinoma recurrence, the method further includes: S5. Machine learning statistical analysis process: A classification model for predicting recurrence of nasopharyngeal carcinoma before treatment is constructed using machine learning methods. The fusion spectra of non-recurrent nasopharyngeal carcinoma patients and the fusion spectra of recurrent nasopharyngeal carcinoma patients are used as training data to train the classification model. Pretreatment plasma samples from the target patient were obtained and converted into cfDNA samples. A fused image of the target patient was acquired using a circularly polarized Raman probe combined with displacement-excited Raman differential spectroscopy. The fused image was then input into a classification model to obtain the classification result. Figure 9 A flowchart illustrating the preparation of a blood cfDNA-silver nanoparticle mixture and its application in circularly polarized SE-SERDS spectroscopy for predicting recurrence before nasopharyngeal carcinoma treatment.

[0056] To further improve the accuracy of nasopharyngeal carcinoma recurrence prediction, three advanced machine learning models (LR, KNN, and SVM) were applied to three SE-SERDS datasets: LHCP alone, RHCP alone, and LHCP+RHCP fusion spectra. Five-fold cross-validation was used, and the confusion matrices of each model under each spectral condition are shown below. Figure 10 As shown in the AI, the corresponding ROC curve is as follows: Figure 10 As shown in JL. For a more intuitive comparison, Figure 11The discrimination performance metrics (sensitivity, specificity, accuracy) for each combination were summarized (including LDA as a baseline comparison): Under LHCP conditions, LDA (83.3%, 80.0%, 81.7%), LR (83.3%, 82.2%, 82.8%), KNN (85.6%, 81.1%, 83.3%), and SVM (86.7%, 83.3%, 85.0%); RHCP Under the given conditions, LDA (90.0%, 87.8%, 88.9%), LR (92.2%, 87.8%, 90.0%), KNN (93.3%, 87.8%, 90.6%), and SVM (93.3%, 88.9%, 91.1%) were observed. Under the LHCP+RHCP conditions, LDA (91.1%, 96.7%, 93.9%), LR (93.3%, 95.6%, 94.4%), KNN (93.3%, 96.7%, 95.0%), and SVM (93.3%, 97.8%, 95.6%) were observed.

[0057] The results reveal three key trends: First, among all algorithms, the LHCP+RHCP fusion with SE-SERDS data performs best, followed by RHCP alone and LHCP SE-SERDS alone; second, for each spectral dataset, SVM outperforms KNN and LR, while LDA performs the worst; third, the performance improvement brought by spectral technique optimization (such as from single polarization to LHCP+RHCP fusion with SE-SERDS) far outweighs the upgrade of machine learning algorithms.

[0058] It is worth noting that the third trend is supported by quantitative evidence: under a unified algorithmic model (such as LDA), optimizing SE-SERDS spectral data from LHCP to RHCP, and further to LHCP+RHCP, improved prediction accuracy by 12.2% (from 81.7% to 93.9%). In contrast, under unified spectral conditions (such as LHCP SE-SERDS), switching from the basic LDA algorithm to SVM only improved accuracy by 3.3% (from 81.7% to 85.0%). This may be because spectral optimization, by capturing more complementary chiral signals from cfDNA, directly enriches the effective information dimensions of the spectrum, fundamentally enhancing the ability to distinguish different prognostic groups. In contrast, algorithm optimization only improves the utilization of existing information, and its performance improvement is limited by the inherent discriminative capabilities of the original spectral data.

[0059] In short, such as Figure 12As shown, the synergistic integration of the SVM model with the LHCP+RHCP fusion SE-SERDS spectroscopy achieved the best performance (accuracy: 95.6%, AUC: 0.991), significantly improving the identification accuracy of nasopharyngeal carcinoma recurrence prediction, providing a robust tool for pre-treatment prognostic assessment, and supporting timely clinical intervention.

[0060] The technical solutions provided in the embodiments of the present invention have at least the following technical effects: 1. This study is the first to explore the integration of circular polarization technology with surface-enhanced displacement-excited Raman differential spectroscopy (SE-SERDS) to obtain high-quality, fluorescence-free circularly polarized surface-enhanced displacement-excited Raman differential spectra. Compared with traditional polynomial fitting methods, it achieves better autofluorescence subtraction and can extract high-quality cfDNA Raman signals, including more subtle peaks related to the prognosis of nasopharyngeal carcinoma. It has great potential for prognostic prediction and guiding personalized treatment in precision oncology.

[0061] 2. Circular polarization excitation improves the discrimination ability by capturing the chiral structural changes of cfDNA. Compared with non-polarization excitation, it produces more significant spectral differences between the relapse and non-relapse groups. More importantly, through signal complementarity integration, the LHCP+RHCP fusion spectrum is superior to the RHCP or LHCP modes alone (accuracy of 93.9% vs. 88.9% and 81.7% under SE-SERDS combined with LDA conditions).

[0062] 3. For the first time, circular polarization technology (including LHCP+RHCP spectral fusion) was integrated with SE-SERDS, and its synergistic advantages significantly improved the accuracy of prognostic prediction for nasopharyngeal carcinoma. This innovative method has great potential as a novel and promising tool for predicting prognostic outcomes before cancer treatment.

[0063] 4. The synergistic integration of LHCP+RHCP with SE-SERDS and SVM models achieved the best prediction accuracy (95.6%). When applied to blood cfDNA analysis, it has the advantages of rapid detection, low cost and high accuracy. It is suitable as a new biosensing tool to predict the recurrence type of nasopharyngeal carcinoma patients, accurately identify high-risk patients and guide individualized treatment.

[0064] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for analyzing blood cfDNA based on circularly polarized surface-enhanced displacement-excited Raman differential spectroscopy, characterized in that, include: Pre-treatment plasma samples from non-recurrent nasopharyngeal carcinoma patients and pre-treatment plasma samples from recurrent nasopharyngeal carcinoma patients were obtained, and cfDNA samples from non-recurrent nasopharyngeal carcinoma patients and cfDNA samples from recurrent nasopharyngeal carcinoma patients were prepared. For each non-recurrent nasopharyngeal carcinoma (NPC) patient's cfDNA sample and each recurrent NPC patient's cfDNA sample, two slightly offset wavelengths were used as the excitation source for a circularly polarized Raman probe. Two Raman differential spectral images at different wavelengths were acquired at the same sample location using the circularly polarized Raman probe. The excitation source was passed through a left-handed circularly polarized filter to obtain an LHCP Raman differential spectral image, and the excitation source was passed through a right-handed circularly polarized filter to obtain an RHCP Raman differential spectral image. The corresponding differential spectra were obtained through iterative calculation using a differential spectral decomposition method, and then reconstructed into background-free LHCP and RHCP spectral images. The background-free LHCP and RHCP spectral images were then fused to obtain a fused image. Principal component analysis is performed on the fused image to reduce its dimensionality to significant principal components, which retain the maximum variance associated with the relapse and non-relapse groups.

2. The method according to claim 1, characterized in that: The method also This includes using linear discriminant analysis to integrate significant principal components and constructing a linear decision boundary based on the distance from the sample to the class centroid, to obtain the LDA posterior probability distribution map of recurrent and non-recurrent nasopharyngeal carcinoma categories.

3. The method according to claim 1, characterized in that: The method also includes using machine learning to construct a classification model for predicting recurrence of nasopharyngeal carcinoma before treatment, and using the fusion spectra of non-recurrent nasopharyngeal carcinoma patients and the fusion spectra of recurrent nasopharyngeal carcinoma patients as training data to train the model. Pretreatment plasma samples from the target patient were obtained and converted into cfDNA samples. A fused image of the target patient was acquired using a circularly polarized Raman probe combined with displacement-excited Raman differential spectroscopy. The fused image was then input into a classification model to obtain the classification result.

4. The method according to claim 1, characterized in that: The LHCP and RHCP spectral images are fused to obtain a fused image. Specifically, the LHCP and RHCP spectral images are normalized separately and then horizontally stitched together to obtain a fused image.

5. The method according to claim 1, characterized in that: The illumination arm of the circularly polarized Raman probe includes a collimator, a bandpass filter, a left-handed polarizer, a right-handed polarizer, and a focusing lens; the collection arm includes a collection lens, a long-pass filter, and a refocusing lens. The collected light is transmitted to a spectrometer for spectral analysis via an optical fiber bundle. The cfDNA spectral acquisition range is 410-1610 cm⁻¹. -1 .

6. The method according to claim 1, characterized in that: The differential spectral decomposition method utilizes two original spectra and their inherent correlation to distinguish Raman signals from interference signals.

7. The method according to claim 1, characterized in that: Plasma cfDNA was extracted from pre-treatment plasma samples of non-recurrent nasopharyngeal carcinoma patients and pre-treatment plasma samples of recurrent nasopharyngeal carcinoma patients, and dissolved in nuclease-free water to obtain cfDNA solutions. The silver nanoparticle mixture was mixed with the cfDNA solution, then dropped onto a substrate and dried to obtain cfDNA samples of non-recurrent nasopharyngeal carcinoma patients and cfDNA samples of recurrent nasopharyngeal carcinoma patients.

Images