A method and system for assisting early diagnosis of head and neck squamous cell carcinoma

Abstract

The present application relates to the technical field of squamous cell carcinoma early diagnosis, and provides a method for assisting early diagnosis of head and neck squamous cell carcinoma, comprising: S1: collecting existing video data obtained by using MET targeted near-infrared fluorescence imaging technology, and training a supervised learning model by using the video data set; S2: building a classifier system, inputting video data to be classified into the classifier system, and applying the trained supervised learning model to segment and determine the pathological properties of the lesion; S3: comparing the efficiency of the built classifier system and human doctors in diagnosing early head and neck squamous cell carcinoma; S4: screening key features related to the occurrence and development of head and neck squamous cell carcinoma through the classifier system, forming a comprehensive data set by connecting the key features with multi-omics data, building an intermediate model based on the comprehensive data set, and performing bioinformatics analysis by using the intermediate model. The present application can improve the early detection rate of HNSCC, improve the accuracy of biopsy, and realize early and accurate tumor diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of early diagnosis of squamous cell carcinoma, and particularly relates to a method and system for assisting early diagnosis of squamous cell carcinoma in head and neck. BACKGROUND

[0002] Head and neck squamous cell carcinoma (HNSCC) mainly refers to squamous cell carcinoma in the nasal cavity, paranasal sinuses, oral cavity, oropharynx, and larynx. Its incidence accounts for about 5% of all tumors in the body, with more than 500,000 new cases worldwide each year (reference: "Cancer statistics, 2023" by Siegel RL, Miller KD, Wagle NS and Jemal A, published in "CA: A Cancer Journal For Clinicians", Vol. 73, No. 1, 2023, and "Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries" by Sung H, Ferlay J, Siegel RL, Laversanne M, SoerjomataramI, Jemal A et al., published in "CA: A Cancer Journal For Clinicians", Vol. 71, No. 3, 2021), and it is one of the major cancers with the fastest rising incidence today (reference: Mody MD, Rocco JW, Yom SS, Haddad RI and Saba NF The current status of HNSCC is unclear, as is the case with the current study, which was published in the journal Lancet (London, England) in Volume 398, Issue 10318, written by Dr. Johnson DE, Burtness B, Leemans CR, Lui VWY, Bauman JE, and Grandis JR in Nat Rev Dis Primers in Volume 6, Issue 1, 2020. Furthermore, although HNSCC is not the most common malignancy, it is one of the top five most lethal malignancies worldwide, with over 380,000 deaths annually (see "Head and neck cancer: Current challenges and future perspectives," written by Dr. Bhat GR, Hyole RG, and Li J, published in Adv Cancer Res in Volume 152, 2021).Despite medical advances and the availability of a range of treatment options (surgery, radiotherapy, chemotherapy, targeted therapy, immunotherapy), the five-year survival rate for HNSCC patients has not improved significantly, according to data reported by Nature. The high local recurrence rate (50-60%) leads to a mortality rate of nearly 50% for HNSCC, with an average five-year survival rate of only about 42% (see The molecular landscape of head and neck cancer by Leemans CR, Snijders PJ, and Brakenhoff RH, Nature Reviews Cancer, Volume 18, Number 5, 2018).

[0003] The low rate of early diagnosis of HNSCC is a major contributor to this predicament. A 2020 review published in the New England Journal of Medicine noted that improving the early diagnosis of HNSCC and the accuracy of preoperative staging are key to prolonging patient survival and improving their quality of life (see "Head and Neck Cancer," authored by Chow LQ, published in the New England Journal of Medicine, Vol. 382, ​​No. 1, 2020). Worldwide, the early diagnosis rate of HNSCC is only 44% to 55% ("Challenges in diagnosing head and neck cancer in primary health care" by Nieminen M, Aro K, Mäkitie A, Harlin V, Kainulainen S, Jouhi L, et al., published in Annals of Medicine, Vol. 53, No. 1, 2021), and the average treatment delay due to misdiagnosis and missed diagnosis is more than 13 months, and more than 60% of the newly diagnosed patients are already in the middle and late stages (referenced by Grafton-Clarke C, Chen KW and Wilcock J, "Diagnosis and referral delays in primary care for oral squamous cell cancer: a systematic review" published in British Journal of General Practice, Vol. 69, No. 679, 2019, and "Clinical Review of the Treatment of Oral Squamous Cell Cancer" by Ligier K, Dejardin O, Launay L, Benoit E, Babin E, Bara S, et al.). et al., "Health professionals and the early detection of head and neck cancers: a population-based study in a high-incidence area," published in BMC Cancer, Vol. 16, No. 1, 2016. Although HNSCC is superficial, early diagnosis is difficult primarily because early cancer is insidious and presents in a variety of ways, making it difficult to distinguish with the naked eye from common benign mucosal lesions such as ulcers and leukoplakia.Epidemiological studies show that the incidence of oral ulcers, oral leukoplakia and other diseases in the population is about 2-4%, of which about 5.9-14.0% progress to squamous cell carcinoma (Li C, Zhang Q, Sun K, Jia H, Shen X, Tang G, et al. Autofluorescence imaging as a noninvasive tool of risk stratification for malignant transformation of oral leukoplakia: A follow-up cohort study. Oral Oncology. 2022;130: 1-9). At the initial stage of this evolution process, it often only shows microscopic cytological changes, and clinicians often cannot accurately biopsy early cancerous tissues based on their experience of visual inspection, which directly leads to clinical misdiagnosis and missed diagnosis. If HNSCC patients can be diagnosed early and treated in time, their 5-year survival rate is expected to increase significantly from the current average of less than 50% to about 85% (Subramaniam N, Balasubramanian D, Murthy S, Kumar N, Vidhyadharan S, Vijayan SN, et al. Predictors of locoregional control in stage I / II oral squamous cell carcinoma classified by AJCC 8th edition. European Journal of Surgical Oncology. 2019;45(11):2126-2130). Therefore, accurately finding small cancer nests at the early stage of HNSCC tumor evolution in time is the key to effective early diagnosis, and is also the bottleneck and breakthrough for ultimately improving patient prognosis and survival time.

[0004] The rapid development of near-infrared fluorescence (NIRF) imaging hardware and related probes has provided a key noninvasive visualization technology for the early diagnosis of HNSCC. NIRF imaging offers high sensitivity, lacks ionizing radiation, is easy to use, and is low-cost, making it particularly suitable for long-term, multiple, real-time imaging. Currently, handheld near-infrared real-time fluorescence imaging devices for HNSCC have gradually matured, and some have entered the clinical trial stage (see “Rapid, non-invasive fluorescence margin assessment: Optical specimen mapping in oral squamous cell carcinoma” by van Keulen S, van den Berg NS, Nishio N, Birkeland A, Zhou Q, Lu G, et al., published in Oral Oncology, Issue 88, 2019; and “Intraoperative tumor assessment using real-time molecular imaging in head and neck cancer patients” by van Keulen S, Nishio N, Fakurnejad S, van den Berg NS, Lu G, Birkeland A, et al., published in Journal of the American College of Surgeons, Volume 229, Issue 6, 2019; and “Fluorescent Molecular Imaging Can Improve Intraoperative Sentinel Imaging” by Krishnan G, van den Berg NS, Nishio N, Kapoor S, Pei J, Freeman L, et al.). Margin Detection in Oral Squamous Cell Carcinoma" was published in the Journal of Nuclear Medicine: Official Publication, Society of Nuclear Medicine, Vol. 63, No. 8, 2022. Near-infrared fluorescence imaging is sensitive to tissue signals within a depth of 5-6 mm. HNSCC originates from squamous cells on the surface of the oral and pharyngeal mucosa, and early lesions are superficial, making it more suitable for near-infrared fluorescence imaging. Furthermore, the addition of targeting elements that respond to tumor-specific molecular markers can further enhance its sensitivity, specificity, and early diagnostic performance.

[0005] Mesenchymal–epithelial transition factor (MET) has been shown to be a driver of various tumors (see "MET-dependent solid tumors—molecular diagnosis and targeted therapy," by Guo R, Luo J, Chang J, Rekhtman N, Arcila M, and Drilon A, published in Nature Reviews Clinical Oncology, 2020). A 2016 study published in Clinical Cancer Research first reported high MET expression in HNSCC ("HGF / Met Signaling in Head and Neck Cancer: Impact on the Tumor Microenvironment," by Hartmann S, Bhola NE, and Grandis JR, published in Clinical Cancer Research, Vol. 22, No. 16, 2016). A Science Signaling interview further pointed out that MET protein is highly expressed in the very early stages of squamous cell carcinoma (see "Science Signaling Podcast for 21 June 2016: MET and skin cancer" by Yuspa SH and VanHook AM, published in the journal Science Signaling in 2016, 9 (433): c14). Previous studies by the applicant's research group also showed that the positive expression rate of MET protein in HNSCC was nearly 80%, which was significantly higher than that in surrounding normal tissues (about 10%) (referenced by Lin B, Wu J, Wang Y, Sun S, Yuan Y, Tao X, et al., "Peptide functionalized upconversion / NIR II luminescent nanoparticles for targeted imaging and therapy of oral squamous cell carcinoma" published in "Biomaterials Science" Volume 9, Issue 3, 2021); through the detection of tongue specimens at serial intervals of the induction time of the spontaneously induced oral squamous cell carcinoma mouse model, we found that MET showed recognizable high expression in the very early stage of HNSCC tumor development.The above results suggest that MET is a key molecular marker for early diagnosis of HNSCC (refer to “Chapter Seven - MET receptor in oncology: From biomarker to therapeutic target” by Malik R, Mambetsariev I, Fricke J, Chawla N, Nam A, Pharaon R, et al. Advances in Cancer Research. Vol. 147. Academic Press; 2020. p. 259-301) and a potential biological target.

[0006] To verify the feasibility of near-infrared fluorescence imaging technology combined with MET-targeting fluorescent probes for early diagnosis of HNSCC, the applicant's research group previously constructed a small-molecule probe cMBP-ICG targeting MET. Through basic and animal experimental research, its effectiveness and safety were confirmed, and a Chinese invention patent has been applied for (Near-infrared fluorescence imaging agent targeting c-Met and use thereof, Patent No.: 202210118240.6). In the same year, the clinical trial "Evaluation of the effectiveness of preoperative topical application of MET-targeting probes to the surface of oral lesions for fluorescence imaging to display lesions" (Registration No.: ChiCTR2200058058) was completed, and it was found that the sensitivity and specificity of the cMBP-ICG probe for displaying tumor lesions were 100% and 75%, respectively, and the accuracy of identifying tumor boundaries increased from 60% based on the experience of experienced surgeons to 83% with the assistance of fluorescence (see "A c-MET-Targeted Topical Fluorescent Probe cMBP-ICG Improves Oral Squamous Cell Carcinoma Detection in Humans" by Wang J, Li S, Wang K, Zhu L, Yang L, Zhu Y, et al. published in Annals of Surgical Oncology 2023, Vol. 30, No. 1). Surface topical fluorescence contrast agents are safe, fast, and convenient, and can help optimize and develop new targeted preparations; they are expected to become non-invasive technologies for diagnosis and follow-up in the future (see "ASO Author Reflections: Topically Applied Targeted Near-Infrared Fluorescence Imaging Probe Helps Detect Head and Neck Squamous Cell Carcinoma" by Wang J, Yuan Y, and Tao X published in Annals of Surgical Oncology 2022).

[0007] Targeted near-infrared fluorescence imaging technology, as a new type of molecular imaging technology, is still in the promotion and learning stage in clinical application (the accuracy of the results of the above-mentioned clinical trials is based on the judgment of near-infrared imaging diagnosis doctors with two years of experience). However, similar to all medical imaging data, the raw video and image data of targeted near-infrared fluorescence imaging contain a large amount of explicit and implicit features. Although the application of artificial intelligence (AI) in near-infrared fluorescence medical imaging is still a blank (referring to the article 'Randomized Clinical Trials of Machine Learning Interventions in Health Care: A Systematic Review' written by Plana D, Shung DL, Grimshaw AA, Saraf A, Sung JJY, and Kann BH published in JAMA Network Open, Volume 5, Issue 9, Article Number e2233946-e2233946), it can be inferred that AI technology can provide strong support for this process. In other medical imaging fields, AI-assisted technology has been proven to be able to enhance image quality, accurately read scan results, speed up the diagnosis process, and reduce diagnostic bias. A large number of studies represented by gastroscopy and cystoscopy show that in the face of video and image information, real-time computer-aided diagnosis systems can provide an additional 20% ~ 25% sensitivity for experienced human physicians and shorten the diagnosis time by nearly 30 minutes per case (referring to the article 'Impact of Artificial Intelligence on Miss Rate of Colorectal Neoplasia' written by Wallace MB, Sharma P, Bhandari P, East J, Antonelli G, Lorenzetti R, et al. published in Gastroenterology, Volume 163, Issue 1, and the article 'An Artificial Intelligence System for the Detection of Bladder Cancer via Cystoscopy: A Multicenter Diagnostic Study' written by Wu S, Chen X, Pan J, Dong W, Diao X, Zhang R, et al. published in Journal of the National Cancer Institute, Volume 114, Issue 2).In the early diagnosis of squamous cell carcinoma, Zhou et al. (Automated detection of tumor regions from oral histological whole slide images using fully convolutional neural networks. Biomedical Signal Processing and Control. 2021;69:102921, written by dos Santos DFD, de Faria PR, Travençolo BAN, do Nascimento MZ) developed a machine learning-based automatic detection of laryngeal squamous cell carcinoma in a bimodal optical imaging microscope, and through the analysis of the test set of pathological images, the AUC of the detection results reached 0.981; Tang et al. (Diagnosis of lymph node metastasis in head and neck squamous cell carcinoma using deep learning. Laryngoscope Investigative Otolaryngology. 2022;7(1), written by Tang H, Li G, Liu C, Huang D, Zhang X, Qiu Y, et al.) used deep convolutional neural networks to extract high-dimensional features from pathological pictures, making the detection accuracy of cervical lymph node HNCSC exceed 97.3%, and improving the accuracy, sensitivity and specificity of the detection results. Lopez-Janeiro et al. (Fully convolutional networks in multimodal nonlinear microscopy images for automated detection of head and neck carcinoma: Pilot study. Head & Neck. 2019;41(1), written by Rodner E, Bocklitz T, von Eggeling F, Ernst G, Chernavskaia O, Popp J, et al.) established a machine learning-based diagnostic model for salivary gland samples, with an accuracy of 84.6% for malignant tumor detection. These methods all use artificial intelligence methods to analyze pathological or imaging data and effectively improve the accuracy of detection.Therefore, we have reason to believe that in the face of more information-rich targeted near-infrared fluorescence imaging data, artificial intelligence has the ability to fully empower human physicians to deeply mine molecular imaging information and further improve the accuracy of early diagnosis of HNSCC, and even reveal more hidden early biological evolution processes of squamous cell carcinoma.

[0008] In summary, MET-targeted near-infrared fluorescence imaging can achieve early visualization of head and neck squamous cell carcinoma and improve its early diagnosis rate; an artificial intelligence assisted diagnosis system is expected to deeply mine the molecular biology information and molecular imaging information provided by this new type of molecular imaging technology, integrate and analyze them, and ultimately affect clinical decision-making. Based on the applicant's previous research, the application will follow the research idea of "development of an artificial intelligence assisted early diagnosis system for head and neck squamous cell carcinoma based on MET-targeted near-infrared fluorescence imaging technology (hereinafter referred to as the targeted diagnosis system) → verification of the targeted diagnosis system → exploration of the biological significance of the key features extracted by the targeted diagnosis system", and carry out the following four parts of research: (1) Data preparation; based on the MET-targeted near-infrared fluorescence imaging video obtained in the previous research, data cleaning and data post-processing such as data segmentation, standardization, and manual outlining of the region of interest are performed, and the lesion paraffin section pathological result is added as a gold standard to prepare data for supervised learning model training. (2) Establishment and verification of the targeted diagnosis system model; a classifier system for automatic segmentation and determination of the pathological nature of the lesion is initially built, model evaluation and hyperparameter tuning are performed through cross-validation between data subsets, learning curves are analyzed, the model is integrated, and the targeted diagnosis system is deployed to the clinical diagnosis and treatment environment for real-time monitoring and debugging of model performance. (3) Comparison of the diagnostic efficiency of the targeted diagnosis system model and human physicians for early head and neck squamous cell carcinoma; under the same test protocol, the diagnostic efficiency of the naked eye group of human physicians, the fluorescence imaging + AI assisted group, the human physician fluorescence imaging assisted group, and the human physician human physician fluorescence imaging + AI assisted group is evaluated, and statistical analysis and result interpretation are performed. (4) Exploration of the biological significance of key features combined with multi-omics data; the key features selected by the targeted diagnosis system and the multi-omics data collected in the previous research are connected, intermediate models are built, and various data integration analyses such as conversion are performed to mine biological information of HNSCC occurrence and development, so that the model's decision can be understood and explained in order to feedback to clinical decision-making. SUMMARY

[0009] In view of the above problems, the purpose of the present application is to provide an auxiliary head and neck squamous cell carcinoma early diagnosis method and system to improve the early detection rate of HNSCC, improve the accuracy of biopsy, realize early and accurate tumor diagnosis, and thus provide a new intelligent visual solution for improving the resection rate and reducing the postoperative recurrence rate. The successful implementation of the present application can also expand the artificial intelligence assisted molecular imaging diagnosis research method to other medical visualization design strategies, and has wide clinical application prospect and basic research value.

[0010] The above application purpose of the present application is realized by the following technical scheme:

[0011] An auxiliary head and neck squamous cell carcinoma early diagnosis method, comprising the following steps:

[0012] S1: A large number of existing video data related to lesions obtained by using MET targeting near-infrared fluorescence imaging technology are pre-recorded as a video data set, the video data set is data cleaned and post-processed, and a supervised learning model with lesion recognition and classification ability is trained using the video data set;

[0013] S2: A classifier system for targeted diagnosis is built as an AI assistant, the classifier system is used to input video data to be classified, the pre-processing and feature extraction of the video data to be classified are performed, and the trained supervised learning model is applied to segment and determine the pathological properties of the lesion;

[0014] S3: The performance of the built classifier system for targeted diagnosis and human doctors in diagnosing early head and neck squamous cell carcinoma is compared;

[0015] S4: The classifier system is used to screen key features related to the occurrence and development of head and neck squamous cell carcinoma, the key features are connected with multi-omics data to form a comprehensive data set, a middle model for processing and analyzing multi-omics data and identifying biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma is built based on the comprehensive data set, and bioinformatics analysis is performed using the middle model to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma.

[0016] Further, in step S1, the video data set is data cleaned and post-processed, and the supervised learning model with lesion recognition and classification ability is trained using the video data set, specifically:

[0017] Each video data in the video data set is data cleaned and post-processed, and each video data is segmented into individual frames, so that each frame of video data is analyzed and processed;

[0018] Performing standardization processing on the video data of each frame, including brightness and contrast adjustment, color standardization, and size standardization, to ensure consistency and comparability of the video data;

[0019] Manually outlining a region of interest related to the lesion in each frame of the video data, and adding a feature label related to the pathological properties to each frame of the video data;

[0020] The supervised learning model is trained by a supervised learning method using the video data to which the feature labels have been added.

[0021] Furthermore, after step S1, the method further includes:

[0022] Paraffin section technology was used to obtain pathological results of lesions, which served as the gold standard for evaluating the performance of the classifier system.

[0023] Furthermore, in step S2, the classifier system for targeted diagnosis is constructed as the AI ​​aid, the video data to be classified is input into the classifier system, preprocessing and feature extraction of the video data to be classified are performed, and the trained supervised learning model is applied to segment and determine the pathological properties of the lesion, specifically:

[0024] Building a preliminary framework of the classifier system, the framework is capable of processing the input video data to be classified, performing preprocessing and feature extraction on the video data to be classified, and applying the trained supervised learning model to segment and determine the pathological properties of the lesions, wherein the preprocessing includes operations such as removing noise and adjusting the brightness, contrast, and color balance of the image, and the feature extraction includes extracting information representing lesion characteristics such as image color, texture, shape, and size;

[0025] Evaluating the performance of the classifier system by comparing the pathological results with the gold standard, and optimizing and adjusting the classifier system according to the results of the performance evaluation to improve the accuracy and reliability of the classifier system;

[0026] Analyze and report the results of the classifier system development and evaluation, including the performance indicators of the classifier system, case analysis of success and failure after segmentation and judgment by the classifier system, and directions for improvement of the classifier system.

[0027] Furthermore, in step S3, the effectiveness of the constructed classifier system for targeted diagnosis is compared with that of human physicians in diagnosing early-stage head and neck squamous cell carcinoma, specifically:

[0028] design a unified test protocol, including specific tasks, evaluation criteria, data sets, and evaluation indicators of the test;

[0029] Set four groups of diagnostic methods including human naked eye group, fluorescence imaging + AI assisted group, human fluorescence imaging assisted group, human fluorescence imaging + AI assisted group, test different diagnostic methods respectively, wherein the human naked eye group is a human doctor only relying on naked eye diagnosis, the fluorescence imaging + AI assisted group is not dependent on human doctors only using fluorescence imaging technology and AI assistance for diagnosis, the human fluorescence imaging assisted group is a human doctor using fluorescence imaging technology for diagnosis, and the human fluorescence imaging + AI assisted group is a human doctor using fluorescence imaging technology and AI assistance for diagnosis;

[0030] According to the designed test protocol, each group is tested, and the diagnostic performance of each group is recorded, and the diagnostic performance of each group is evaluated by using the predetermined evaluation criteria including accuracy, recall rate, and FI score;

[0031] Statistical analysis is performed to compare the diagnostic performance of different groups, including using ANOVA, t-test and other methods to determine whether the performance difference is significant, and explaining the statistical analysis results, including the advantages and disadvantages of each group, and whether fluorescence imaging technology and AI assistance can significantly improve the diagnostic performance.

[0032] Further, in step S4, the classifier system is used to screen key features related to the occurrence and development of head and neck squamous cell carcinoma, the key features are combined with multi-omics data to form a comprehensive data set, and an intermediate model for processing and analyzing multi-omics data and identifying biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma is built based on the comprehensive data set, specifically:

[0033] The classifier system is used to screen the key features related to the occurrence and development of head and neck squamous cell carcinoma;

[0034] Collect multi-omics data including genomics, transcriptomics, proteomics, and metabolomics;

[0035] The screened key features are combined with the collected multi-omics data to form the comprehensive data set, and the intermediate model is built based on the comprehensive data set, which can process and analyze the multi-omics data, and identify biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma;

[0036] The output of the intermediate model is converted, including converting the expression levels or variation data of biomarkers into a more easily understood and used form, to facilitate analysis and interpretation.

[0037] Further, in step S4, bioinformatics analysis is performed using the intermediate model to explore the biological basis and therapeutic targets of the head and neck squamous cell carcinoma, specifically:

[0038] Bioinformatics analysis is performed using the intermediate model to explore the biological basis and therapeutic targets of the head and neck squamous cell carcinoma;

[0039] The interpretability of the intermediate model is analyzed and evaluated to ensure that the decision-making process of the intermediate model can be understood and interpreted, including using model interpretation techniques such as LIME or SHAP to understand how the intermediate model uses input data to make decisions;

[0040] The interpretability results of the intermediate model and the findings of bioinformatics analysis are fed back to clinical decision-making to provide better patient management and treatment options, and the model and analysis method are continuously optimized and updated according to clinical feedback and new research findings to maintain relevance and accuracy in the study and clinical application of the head and neck squamous cell carcinoma.

[0041] An auxiliary head and neck squamous cell carcinoma early diagnosis system for performing the auxiliary head and neck squamous cell carcinoma early diagnosis method as described above, comprising:

[0042] A data preparation module for pre-including a large number of existing video data related to lesions obtained using MET targeting near-infrared fluorescence imaging technology as a video data set, performing data cleaning and post-processing on the video data set, and training a supervised learning model with lesion recognition and classification capabilities using the video data set;

[0043] A targeted diagnosis system establishment module for building a classifier system for targeted diagnosis as AI assistance, inputting video data to be classified into the classifier system, performing preprocessing and feature extraction of the video data to be classified, and applying the trained supervised learning model to segment and determine the pathological properties of the lesion;

[0044] A diagnosis performance comparison module for comparing the performance of the classifier system built for targeted diagnosis and the performance of human physicians in diagnosing early head and neck squamous cell carcinoma;

[0045] A biological significance exploration module is used to screen key features related to the occurrence and development of head and neck squamous cell carcinoma through the classifier system, concatenate the key features with multi-omics data to form a comprehensive data set, build an intermediate model for processing and analyzing multi-omics data based on the comprehensive data set, and identify biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma, and use the intermediate model to perform bioinformatics analysis to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma.

[0046] A computer device includes a memory and one or more processors, wherein the memory stores computer code, and when the computer code is executed by the one or more processors, the one or more processors execute the above method.

[0047] A computer-readable storage medium stores computer code. When the computer code is executed, the above method is performed.

[0048] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0049] (1) The present invention uses MET targeted probes combined with near-infrared fluorescence imaging technology as the data source for machine learning. This method utilizes the high sensitivity and deep tissue penetration of near-infrared fluorescence imaging, combined with the high specificity of the MET probe, to more accurately mark and identify the lesion area of ​​head and neck squamous cell carcinoma (HNSCC). This technology provides a high-quality training data set with richer information levels for machine learning algorithms, which greatly enhances the accuracy of disease identification and classification. With this data, the classifier system as an AI model can be trained to obtain the precise boundaries and biological characteristics of the lesions, thereby providing a new perspective for the early diagnosis of HNSCC. In addition, matching these image data with high-quality pathological results and immunohistochemistry results provides a "gold standard" for verifying the accuracy of the machine learning model, further enhancing the value and reliability of the AI-assisted system in actual clinical applications. This multimodal data fusion strategy not only optimizes the existing diagnostic process, but also enhances the interpretability of the model, providing doctors with an interpretable diagnostic basis, and reflects the application potential of artificial intelligence in the field of precision medicine.

[0050] (2) By analyzing multi-omics data related to MET, including gene expression, protein expression, and other molecular changes, we can identify biological events that influence the course of HNSCC and use these events to build and iteratively design more accurate artificial intelligence models. This approach can distill key information in the disease mechanism, providing dynamic, continuous biomarker assessment for early diagnosis and treatment strategy development. Through continuous model iteration, it is possible to further achieve prediction of targeted drug treatment response, thereby achieving individualized treatment. In clinical decision-making, this research feature allows doctors to obtain data-driven insights that are automatically learned from large-scale data sets by deep learning algorithms. This iterative design method ensures the timeliness and adaptability of the model, enabling it to keep pace with the latest research findings and clinical practice. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 Flowchart of the overall process of the application for assisting in early diagnosis of head and neck squamous cell carcinoma;

[0052] Figure 2 Flowchart of the overall design route of the application for assisting in early diagnosis of head and neck squamous cell carcinoma;

[0053] Figure 3 Flowchart of the data processing of the application;

[0054] Figure 4 Flowchart of the development and verification of the supervised learning model as an AI model of the application;

[0055] Figure 5 Overall structure diagram of the application for assisting in early diagnosis of head and neck squamous cell carcinoma. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0057] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "a," "an," and "the" as used herein can include plural forms. It should be further understood that the use of the term "including" in the specification of the present application means that the stated features, integers, steps, operations, elements, and / or components exist, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0058] Head and neck squamous cell carcinoma (HNSCC) accounts for approximately 90% of head and neck malignancies. Its insidious nature and low rate of early diagnosis contribute to its high recurrence rate and poor prognosis. Targeted near-infrared fluorescence imaging offers new possibilities for improving early tumor imaging. Initial clinical studies have demonstrated that a small molecule probe targeting MET has increased its accuracy in detecting early oral mucosal cancer from 56% to approximately 84%.

[0059] By utilizing MET-targeted near-infrared fluorescence imaging technology, early visual diagnosis of head and neck squamous cell carcinoma can be achieved, thereby significantly improving the early diagnosis rate of the disease. With the introduction of artificial intelligence-assisted diagnosis systems, it is expected that the rich molecular biology and molecular imaging information provided by this new molecular imaging technology will be deeply mined and analyzed. By integrating and analyzing this information, we can better understand the biological mechanisms of the lesions and thus influence clinical decision-making. This project plans to carry out the following three major studies based on the applicant's previous research results:

[0060] (1) Establishment and verification of targeted diagnosis system model: Introduce machine learning to conduct in-depth mining of the above molecular imaging data and establish a targeted diagnosis classifier system.

[0061] Initially build a classifier system for automatically segmenting and determining the pathological properties of lesions. Evaluate model performance on different data subsets through cross-validation and perform hyperparameter tuning. Analyze learning curves to identify potential overfitting or underfitting issues and take appropriate measures to address them. Perform model ensembles to improve the stability and accuracy of the classifier system. Deploy the targeted diagnostic system in a clinical setting, and monitor and debug model performance in real time to ensure accuracy and reliability.

[0062] (2) Diagnostic efficacy ratio: The effectiveness of the targeted diagnostic model is verified by comparing it with the diagnosis of human physicians.

[0063] Design a unified testing protocol to evaluate the efficacy of different diagnostic methods (human physician naked eye diagnosis, fluorescence imaging + AI-assisted diagnosis, fluorescence imaging-assisted diagnosis, and fluorescence imaging + AI-assisted diagnosis) for the early diagnosis of head and neck squamous cell carcinoma. Statistical analysis will be used to compare the diagnostic efficacy of different diagnostic methods, and the results will be interpreted and discussed.

[0064] (3) Combine multi-omics data to explore the biological significance of key features: Combine existing clinical data, imaging data, pathological data and other multi-omics data to explore the key features suggested by the targeted diagnostic model and mine the biological information of the occurrence and development of HNSCC; ultimately, the model's decisions can be understood and explained so that they can be fed back into clinical decision-making.

[0065] The key features screened by the targeted diagnostic system are integrated with the multi-omics data (including genomics, transcriptomics, proteomics, and metabolomics data) collected in the previous studies. An intermediate model is constructed to analyze and interpret the integrated data to mine the bioinformatics basis of HNSCC occurrence and development. The biological significance of the key features identified by the model is explored, and these findings are fed back to clinical decision-making to provide more sound clinical diagnosis and treatment options.

[0066] The present application introduces artificial intelligence into the field of molecular imaging, which is expected to improve the early detection rate and biopsy accuracy of HNSCC, achieve early precise tumor diagnosis, and thus provide new intelligent diagnosis and treatment options for improving surgical resection rate and reducing postoperative recurrence rate. At the same time, the successful implementation of the present application is also expected to expand this innovative research approach to other medical visualization design strategies, which has a wide clinical application prospect and basic research value. At the social medical level, the smooth implementation of this project will not only promote the progress of early diagnosis technology for head and neck squamous cell carcinoma, but also is expected to provide strong decision support for clinicians, and ultimately improve the diagnosis and treatment effect and life quality of patients.

[0067] The following is illustrated by specific embodiments:

[0068] First embodiment

[0069] As shown in Figure 1 and 2 , the present embodiment provides a method for assisting early diagnosis of head and neck squamous cell carcinoma, comprising the following steps:

[0070] S1: A large number of existing video data related to lesions obtained by using MET targeted near-infrared fluorescence imaging technology are pre-recorded as a video data set, the video data set is data cleaned and post-processed, and a supervised learning model with lesion recognition and classification ability is trained using the video data set.

[0071] As described in Figure 3 , the present embodiment pre-recording a large number of existing video data related to lesions obtained by using MET targeted near-infrared fluorescence imaging technology as a video data set, screening and excluding all NIRF format video data in the video data set, and performing video quality control, the video data passing the quality control is included in the video data set.

[0072] After the video data set is determined, the video data set is data cleaned and post-processed, and a supervised learning model with lesion recognition and classification ability is trained using the video data set, specifically:

[0073] Each of the video data in the set is individually data cleaned and post-processed, and each of the video data is segmented into individual frames for analysis and processing of the video data of each frame. Video processing software or programming libraries such as OpenCV can be used to read video files and save them frame by frame as image files.

[0074] The video data of each frame is standardized, including brightness and contrast adjustment, color normalization, and size normalization, to ensure consistency and comparability of the video data.

[0075] (1) Brightness and contrast adjustment:

[0076] The average brightness value of the image can be calculated, and then the brightness can be adjusted to a specific range as needed. For example, the contrast of the image can be enhanced by linear transformation or histogram equalization to make the details in the image clearer.

[0077] Using image enhancement algorithms such as adaptive histogram equalization (AHE) or contrast limited adaptive histogram equalization (CLAHE), local contrast can be enhanced without changing the overall brightness.

[0078] (2) Color normalization:

[0079] If the image is colored, the color space can be converted to a standard color space such as RGB or HSV. Then, the color channels are standardized, for example, adjusting the color values to a specific range or mean zero, variance one normal distribution.

[0080] For multiple frames of images, the average color value of all images can be calculated, and the color of each frame of image can be adjusted to be close to this average value to ensure the consistency of the color.

[0081] (3) Size normalization:

[0082] If the sizes of the images are inconsistent, they can be adjusted to the same size. This can be achieved by image scaling algorithms such as bilinear interpolation or nearest neighbor interpolation. Ensure that the proportions and clarity of the image are maintained when adjusting the size.

[0083] Manually outline the region of interest related to the lesion in each frame of the video data, and add feature labels related to pathological properties to each frame of the video data.

[0084] There are many image annotation tools available, such as LabelImg, VGG Image Annotator (VIA), etc. These tools allow users to manually draw rectangles, polygons, or other shapes on images to mark regions of interest. Annotators need to have some expertise in pathology to accurately identify and outline areas related to lesions. To ensure the accuracy of annotations, quality control checks can be performed. Multiple annotators can be assigned to annotate the same set of images, and their results can be compared to determine the consistency of annotations. If there are discrepancies, discussions and corrections can be made. Annotators should be regularly trained and evaluated to improve the quality and accuracy of annotations. Professional pathologists or doctors are needed to determine appropriate feature labels based on the characteristics and pathological nature of lesions. For example, based on the morphology, color, size, edge, and other characteristics of lesions, they can be labeled as benign, malignant, inflammatory, and other different pathological states. To ensure consistency in labeling, clear annotation guidelines and standards can be established. Annotators should follow these guidelines when adding labels to avoid subjective differences. Labeling can be reviewed and verified, for example, by another professional checking the annotated images to ensure the accuracy of the labels.

[0085] As shown in Figure 4 The supervised learning model is trained using the video data with the added feature labels. The supervised learning model, which is the AI model used later, is trained by randomly dividing the video data set into a training set and a validation set during training. The model is trained using the training set and validated using the validation set.

[0086] After step S1, it also includes using paraffin section technology to obtain the pathological results of the lesions as the gold standard for evaluating the performance of the classifier system.

[0087] Paraffin section is a commonly used histopathological examination method. First, a tissue sample containing lesions is obtained from the patient's body. Then, the tissue sample is subjected to a series of processes, including fixation (usually using formalin or other fixatives to keep cells and structures in the tissue stable), dehydration (removing water from the tissue to facilitate subsequent immersion in paraffin), transparency (using organic solvents to make the tissue transparent to facilitate the penetration of paraffin), wax immersion (immersing the tissue in liquid paraffin to allow paraffin to penetrate the tissue), and embedding (placing the wax-immersed tissue into a mold, pouring liquid paraffin into the mold, and allowing the paraffin to cool and solidify to form a wax block containing the tissue).

[0088] Next, the wax block is cut into extremely thin sections (usually several microns thick) using a microtome. These sections can be attached to glass slides and stained (such as hematoxylin-eosin staining) to facilitate observation of the cellular morphology, structure, and pathological changes of the tissue under a microscope.

[0089] A pathologist observes the morphology of the tissue on a paraffin section under a microscope and diagnoses the lesion based on indicators such as the shape, size, arrangement of cells, characteristics of the nucleus, and the presence or absence of abnormal cells. For example, if the presence of cancer cells, obvious cellular atypia, and disordered tissue structure are observed, it may be diagnosed as a malignant tumor; if the cell morphology is relatively normal, there is no obvious atypia and abnormal proliferation, it may be diagnosed as a benign lesion or normal tissue.

[0090] Since the pathological result of the paraffin section is obtained by professional pathologists observing and diagnosing the tissue directly, it has high accuracy and reliability, and is therefore used as the gold standard for the classifier system. This means that when evaluating the performance of the classifier system, its predicted results are compared with the pathological results of the paraffin section.

[0091] S2: Build a classifier system for targeted diagnosis as an AI assistant, input the video data to be classified into the classifier system, perform preprocessing and feature extraction of the video data to be classified, and apply the trained supervised learning model to segment and determine the pathological properties of the lesion.

[0092] Build a preliminary framework of the classifier system, which can process the input video data to be classified, perform preprocessing and feature extraction on the video data to be classified, and apply the trained supervised learning model to segment and determine the pathological properties of the lesion, wherein the preprocessing includes operations such as removing noise, adjusting the brightness, contrast and color balance of the image, and the feature extraction includes extraction of information representing the characteristics of the lesion such as color, texture, shape and size.

[0093] Develop and integrate an automatic segmentation algorithm to identify and segment the lesion area in the image without human intervention. Use the trained model to automatically determine the pathological properties of the lesion, such as benign or malignant.

[0094] Compare the performance of the classifier system with the pathological results of the gold standard, and optimize and adjust the classifier system according to the results of the performance evaluation to provide the accuracy and reliability of the classifier system.

[0095] Analyze and report the results of the development and evaluation of the classifier system, including the performance indicators of the classifier system, case analysis of success and failure after segmentation and judgment by the classifier system, and the direction in which the classifier system can be improved.

[0096] For example:

[0097] (1) Performance indicators of the classifier:

[0098] Accuracy: A measure of a classifier's ability to correctly identify the pathological nature of a lesion. It is typically calculated by dividing the number of correctly classified samples by the total number of samples. For example, if a classifier correctly identifies 85 of 100 lesion samples, the accuracy is 85%.

[0099] Sensitivity: Also known as recall, this measure measures a classifier's ability to correctly identify positive samples (e.g., malignant lesions). High sensitivity means the classifier is better able to detect true positive cases, reducing the likelihood of missed diagnoses.

[0100] Specificity: This measures the classifier's ability to correctly identify negative samples (e.g., benign lesions). A high specificity means the classifier is better able to exclude true negative cases, reducing the possibility of misdiagnosis.

[0101] Precision: This reflects how precisely the classifier determines a lesion. For example, if the classifier determines that a lesion is malignant but also specifies its possible subtype or grade, then the precision is high.

[0102] Other indicators: May also include F1 value (an indicator that comprehensively considers accuracy and sensitivity), area under the ROC curve, etc.

[0103] (2) Case analysis of success and failure:

[0104] Success Cases: Analyze selected cases where the classifier system accurately determined the pathological nature of lesions. For example, in a complex lesion image, the classifier correctly determined it to be malignant, consistent with the gold standard pathology results. Analyzing these success cases can help us summarize the advantages of the classifier system and effective methods, such as specific image features and algorithmic advantages.

[0105] Failure Cases: Analyze cases where the classifier system misjudged a patient. For example, the classifier might mistakenly classify a benign lesion as malignant, or fail to detect a small malignant lesion. By analyzing failure cases, we can identify system deficiencies, such as incomplete feature extraction and algorithmic limitations.

[0106] (3) Possible improvement directions:

[0107] Algorithm improvement: Based on performance indicators and case analysis, propose improvements to the classifier algorithm. For example, adopt more advanced machine learning algorithms, optimize feature extraction methods, adjust parameters, etc.

[0108] Data augmentation: If you find that the classifier system performs poorly on certain types of lesions, you can consider adding relevant types of training data to improve the system's generalization ability.

[0109] Combining with other technologies: Explore the possibility of combining with other technologies to improve the performance of the classifier, such as combining with other imaging technologies, introducing clinical information, etc.

[0110] User interface optimization: If the classifier system is designed for use by clinicians, one can also consider optimizing the user interface to make it easier to operate and understand.

[0111] S3: The effectiveness of the classifier system constructed for targeted diagnosis was compared with that of human physicians in diagnosing early-stage head and neck squamous cell carcinoma.

[0112] In this embodiment, step S3 is specifically as follows:

[0113] Design a unified testing protocol, including specific test tasks, evaluation criteria, data sets, and evaluation metrics. The following are specific examples:

[0114] (1) Specific tasks of the test:

[0115] Clarify the goal of the test, such as comparing the diagnostic efficacy of different diagnostic methods (naked eye diagnosis, fluorescence imaging-assisted diagnosis, AI-assisted diagnosis, etc.) for early head and neck squamous cell carcinoma.

[0116] Determine the specific operational procedures for the test, such as having a doctor or model diagnose a given set of images and record the diagnosis results.

[0117] (2) Evaluation criteria:

[0118] Define the criteria for judging the correctness of the diagnosis. For example, use the pathological results of the paraffin section of the lesion as the gold standard and compare them with the results of different diagnostic methods.

[0119] It is possible to consider setting different levels of accuracy requirements. For example, for the diagnosis of malignant lesions, the accuracy rate must reach a certain level.

[0120] (3) Dataset:

[0121] Select a dataset for testing. This dataset should be representative and cover early-stage head and neck squamous cell carcinoma cases of different types and severities.

[0122] The dataset can come from multiple sources, such as clinical records in hospitals, medical research databases, etc. Ensure the image quality and annotation accuracy in the dataset.

[0123] (4) Evaluation indicators:

[0124] Determine the metrics used to measure the effectiveness of different diagnostic methods. Common evaluation metrics include precision, recall, F1 score, etc.

[0125] Accuracy measures the overall correctness of the diagnostic results; recall measures the proportion of correctly identified positive cases (e.g., malignant lesions); F1 score considers both accuracy and recall.

[0126] Other indicators such as specificity (the proportion of correctly identified negative cases) and precision can also be selected based on specific needs.

[0127] Four groups of diagnostic methods are set up, including the naked eye group of human doctors, the fluorescence imaging + AI auxiliary group, the fluorescence imaging auxiliary group of human doctors, and the fluorescence imaging + AI auxiliary group of human doctors. Different diagnostic methods are tested, wherein the naked eye group of human doctors is diagnosed by human doctors only with naked eyes, the fluorescence imaging + AI auxiliary group is diagnosed by using fluorescence imaging technology and AI assistance without relying on human doctors, the fluorescence imaging auxiliary group of human doctors is diagnosed by using fluorescence imaging technology, and the fluorescence imaging + AI auxiliary group of human doctors is diagnosed by using fluorescence imaging technology and AI assistance.

[0128] According to the designed test protocol, each group is tested, and the diagnostic performance of each group is recorded. The diagnostic performance of each group is evaluated using pre-determined evaluation criteria including accuracy, recall, and F1 score.

[0129] Statistical analysis is performed to compare the diagnostic performance of different groups, including using ANOVA, t-test, etc. to determine whether the performance difference is significant. The statistical analysis results are explained, including the advantages and disadvantages of each group, and whether fluorescence imaging technology and AI assistance can significantly improve the diagnostic performance.

[0130] ANOVA (Analysis of Variance): When comparing the diagnostic performance of multiple groups, ANOVA can be used. ANOVA can test whether the means of multiple groups are significantly different. If the ANOVA result shows that the difference between groups is significant, further post-hoc test can be performed to determine which specific groups have differences. For example, through ANOVA, it can be determined whether the accuracy, recall, and other indicators under different diagnostic methods are significantly different.

[0131] t-test: If comparing the diagnostic performance of two groups, such as the naked eye group and the fluorescence imaging auxiliary group, t-test can be used. t-test can test whether the means of two samples are significantly different. Similarly, according to the results of t-test, it can be determined whether the two diagnostic methods have significant differences in specific indicators.

[0132] S4: screening the key features related to the occurrence and development of head and neck squamous cell carcinoma through the classifier system, concatenating the key features with multi-omics data to form a comprehensive data set, building an intermediate model for processing and analyzing multi-omics data based on the comprehensive data set, and identifying biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma, using the intermediate model for bioinformatics analysis to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma.

[0133] In this embodiment, step S4, specifically:

[0134] The key features related to the occurrence and development of head and neck squamous cell carcinoma are screened through the classifier system;

[0135] Collecting multi-omics data including genomics, transcriptomics, proteomics and metabolomics;

[0136] The key features screened are concatenated with the collected multi-omics data to form the comprehensive data set, and the intermediate model is built based on the comprehensive data set, which can process and analyze the multi-omics data, and identify biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma;

[0137] The output of the intermediate model is converted, including converting the expression level or variation data of biomarkers into a more easily understood and used form, to facilitate analysis and interpretation.

[0138] Bioinformatics analysis is performed using the intermediate model to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma;

[0139] The interpretability of the intermediate model is analyzed and evaluated to ensure that the decision-making process of the intermediate model can be understood and interpreted, including using model interpretation techniques such as LIME or SHAP to understand how the intermediate model uses input data to make decisions;

[0140] The interpretability results of the intermediate model and the findings of bioinformatics analysis are fed back to clinical decision-making to provide better patient management and treatment options, and the model and analysis method are continuously optimized and updated according to clinical feedback and new research findings to maintain relevance and accuracy in head and neck squamous cell carcinoma research and clinical applications.

[0141] Second embodiment

[0142] As Figure 5As shown, this embodiment provides an auxiliary head and neck squamous cell carcinoma early diagnosis system for executing the auxiliary head and neck squamous cell carcinoma early diagnosis method as described in the first embodiment, comprising:

[0143] Data preparation module 1 is used to pre-collect a large amount of existing video data related to lesions obtained using MET-targeted near-infrared fluorescence imaging technology as a video data set, perform data cleaning and post-processing on the video data set, and use the video data set to train a supervised learning model capable of identifying and classifying lesions;

[0144] Targeted diagnosis system establishment module 2 is used to build a classifier system for targeted diagnosis as an AI aid, using the classifier system to input video data to be classified, perform preprocessing and feature extraction of the video data to be classified, and apply the trained supervised learning model to segment and determine the pathological nature of the lesion;

[0145] A diagnostic efficacy comparison module 3 is used to compare the efficacy of the constructed classifier system for targeted diagnosis with that of human physicians in diagnosing early-stage head and neck squamous cell carcinoma;

[0146] Biological significance exploration module 4 is used to screen key features related to the occurrence and development of head and neck squamous cell carcinoma through the classifier system, concatenate the key features with multi-omics data to form a comprehensive data set, build an intermediate model for processing and analyzing multi-omics data based on the comprehensive data set, and identify biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma, and use the intermediate model to perform bioinformatics analysis to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma.

[0147] A computer-readable storage medium stores computer code. When the computer code is executed, the above-described method is performed. Those skilled in the art will appreciate that all or part of the steps in the various methods of the above-described embodiments can be performed by a program instructing related hardware. The program can be stored in a computer-readable storage medium. The storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0148] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

[0149] The technical features of the above-described embodiments can be combined in any manner. For the sake of brevity, not all possible combinations of the technical features in the above-described embodiments are described, however, as long as the combinations of the technical features do not contradict each other, they should be considered to be within the scope of the present specification.

[0150] It should be noted that the above-described embodiments can be freely combined as needed. The above-described only is the preferred embodiments of the present application, it should be noted that, for those skilled in the art, without departing from the principles of the present application, can make a number of improvements and refinements, these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for assisting the early diagnosis of head and neck squamous cell carcinoma, characterized in that: The following steps are involved: S1: Pre-collect a large amount of existing video data related to lesions obtained using MET-targeted near-infrared fluorescence imaging technology as a video data set, perform data cleaning and post-processing on the video data set, and use the video data set to train a supervised learning model capable of identifying and classifying lesions; S2: Building a classifier system for targeted diagnosis as an AI aid, using the classifier system to input video data to be classified, performing preprocessing and feature extraction on the video data to be classified, and applying the trained supervised learning model to segment and determine the pathological nature of the lesion; S3: Comparing the effectiveness of the classifier system constructed for targeted diagnosis with that of human physicians in diagnosing early-stage head and neck squamous cell carcinoma; S4: Using the classifier system to screen key features associated with the occurrence and progression of head and neck squamous cell carcinoma, the key features are concatenated with multi-omics data to form a comprehensive dataset, and based on the comprehensive dataset, an intermediate model is constructed for processing and analyzing multi-omics data, as well as identifying biomarkers and model pathways associated with the occurrence and progression of head and neck squamous cell carcinoma. The intermediate model is used to perform bioinformatics analysis to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma; After step S1, the method further includes: using paraffin section technology to obtain pathological results of the lesion as a gold standard for evaluating the performance of the classifier system; In step S2, the classifier system for targeted diagnosis is constructed as the AI ​​aid, the classifier system is used to input the video data to be classified, preprocessing and feature extraction of the video data to be classified are performed, and the trained supervised learning model is applied to segment and determine the pathological properties of the lesion, specifically: Building a preliminary framework of the classifier system, the framework is capable of processing the input video data to be classified, performing preprocessing and feature extraction on the video data to be classified, and applying the trained supervised learning model to segment and determine the pathological properties of the lesions, wherein the preprocessing includes operations such as removing noise and adjusting the brightness, contrast, and color balance of the image, and the feature extraction includes extracting information representing lesion characteristics such as image color, texture, shape, and size; Evaluating the performance of the classifier system by comparing the pathological results with the gold standard, and optimizing and adjusting the classifier system according to the results of the performance evaluation to improve the accuracy and reliability of the classifier system; Analyze and report the results of the classifier system development and evaluation, including the performance indicators of the classifier system, case analysis of success and failure after segmentation and judgment by the classifier system, and directions for improvement of the classifier system.

2. The method for assisting early diagnosis of head and neck squamous cell carcinoma according to claim 1, characterized in that: In step S1, the video data set is cleaned and post-processed, and the supervised learning model capable of identifying and classifying lesions is trained using the video data set, specifically: Performing data cleaning and post-processing on each of the video data in the video data set, and dividing each of the video data into separate frames, so as to analyze and process the video data of each frame; Performing standardization processing on the video data of each frame, including brightness and contrast adjustment, color standardization, and size standardization, to ensure consistency and comparability of the video data; Manually outlining a region of interest related to the lesion in each frame of the video data, and adding a feature label related to the pathological properties to each frame of the video data; The supervised learning model is trained by a supervised learning method using the video data to which the feature labels have been added.

3. The method for assisting early diagnosis of head and neck squamous cell carcinoma according to claim 1, characterized in that: In step S3, the effectiveness of the constructed classifier system for targeted diagnosis is compared with that of human physicians in diagnosing early-stage head and neck squamous cell carcinoma, specifically: Design a unified testing protocol, including specific test tasks, evaluation criteria, data sets, and evaluation metrics; Four diagnostic methods were set up, including a human physician naked eye group, a fluorescence imaging + AI-assisted group, a human physician fluorescence imaging-assisted group, and a human physician fluorescence imaging + AI-assisted group. Different diagnostic methods were tested separately. The human physician naked eye group was a human physician who made a diagnosis solely with the naked eye; the fluorescence imaging + AI-assisted group was a human physician who made a diagnosis without relying on a human physician and only used fluorescence imaging technology and AI assistance; the human physician fluorescence imaging-assisted group was a human physician who made a diagnosis using fluorescence imaging technology; and the human physician fluorescence imaging + AI-assisted group was a human physician who made a diagnosis with the assistance of fluorescence imaging technology and AI. According to the designed test protocol, each group is tested, and the diagnostic performance of each group is recorded, and the diagnostic performance of each group is evaluated using the predetermined evaluation criteria including accuracy, recall rate, and FI score; Perform statistical analysis to compare the diagnostic efficacy between different groups, including using analysis of variance (ANOVA) and t-test to determine whether the efficacy differences are significant, interpret the statistical analysis results, including the advantages and disadvantages of each group, and whether fluorescence imaging technology and AI assistance can significantly improve diagnostic efficacy.

4. The method for assisting early diagnosis of head and neck squamous cell carcinoma according to claim 1, characterized in that: In step S4, the classifier system is used to screen key features related to the occurrence and progression of head and neck squamous cell carcinoma, and the key features are connected in series with the multi-omics data to form a comprehensive data set. Based on the comprehensive data set, an intermediate model for processing and analyzing the multi-omics data and identifying biomarkers and model pathways related to the occurrence and progression of head and neck squamous cell carcinoma is constructed, specifically: Screening out the key features related to the occurrence and development of head and neck squamous cell carcinoma through the classifier system; Collect multi-omics data including genomics, transcriptomics, proteomics, and metabolomics; Concatenating the screened key features with the collected multi-omics data to form the comprehensive data set, and building the intermediate model based on the comprehensive data set, wherein the intermediate model is capable of processing and analyzing the multi-omics data, and identifying biomarkers and model pathways related to the occurrence and progression of head and neck squamous cell carcinoma; The output of the intermediate model is transformed to facilitate analysis and interpretation, including converting the expression level or variation data of the biomarker into a form that is more easily understood and used.

5. The method for assisting early diagnosis of head and neck squamous cell carcinoma according to claim 1, characterized in that: In step S4, bioinformatics analysis is performed using the intermediate model to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma, specifically: performing bioinformatics analysis using the intermediate model to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma; Analyze and evaluate the interpretability of the intermediate model to ensure that the decision-making process of the intermediate model can be understood and explained, including using model interpretation techniques such as LIME or SHAP to understand how the intermediate model uses input data to make decisions; Feed the interpretative results of the intermediate model and the findings of bioinformatics analysis back into clinical decision-making to provide better patient management and treatment options, and continuously optimize and update the model and analysis methods based on clinical feedback and new research findings to maintain relevance and accuracy in head and neck squamous cell carcinoma research and clinical applications.

6. An assisted early diagnosis system for head and neck squamous cell carcinoma for executing the assisted early diagnosis method for head and neck squamous cell carcinoma according to any one of claims 1 to 5, characterized in that: include: A data preparation module is used to pre-collect a large amount of existing video data related to lesions obtained using MET-targeted near-infrared fluorescence imaging technology as a video data set, perform data cleaning and post-processing on the video data set, and use the video data set to train a supervised learning model capable of identifying and classifying lesions; A targeted diagnosis system establishment module is used to build a classifier system for targeted diagnosis as an AI aid, using the classifier system to input video data to be classified, perform preprocessing and feature extraction on the video data to be classified, and apply the trained supervised learning model to segment and determine the pathological nature of the lesion; A diagnostic efficacy comparison module, for comparing the efficacy of the classifier system constructed for targeted diagnosis with that of human physicians in diagnosing early-stage head and neck squamous cell carcinoma; A biological significance exploration module is used to screen key features related to the occurrence and development of head and neck squamous cell carcinoma through the classifier system, concatenate the key features with multi-omics data to form a comprehensive data set, build an intermediate model for processing and analyzing multi-omics data based on the comprehensive data set, and identify biomarkers and model pathways related to the occurrence and development of head and neck squamous cell carcinoma, and use the intermediate model to perform bioinformatics analysis to explore the biological basis and therapeutic targets of head and neck squamous cell carcinoma.

7. A computer device comprising a memory and one or more processors, wherein the memory stores computer code, and when the computer code is executed by the one or more processors, the one or more processors are caused to perform the method according to any one of claims 1 to 5. 8 . A computer-readable storage medium storing computer code, wherein when the computer code is executed, the method according to claim 1 is performed.

Images