Study On Artificial Intelligence Recognition And Application Of Lymphovascular Invasion In Hypopharyngeal Cancer Based On Pathological Sections

Background:Head and neck squamous cell carcinoma(HNSCC)is a common malignant tumor,and hypopharyngeal squamous cell carcinoma is an important subtype of HNSCC,accounting for about 3%to 5%of HNSCC.Due to the special anatomy of the hypopharynx and the insidious onset of hypopharyngeal cancer,patients often present at an advanced stage with local or distant metastasis,resulting in poor prognosis.Lymphovascular invasion is one of the important pathways for malignant tumor growth and spread.Lymphovascular invasion is closely related to the prognosis of tumor patients.Currently,the diagnosis of lymphovascular invasion mainly relies on postoperative routine pathology,which is time-consuming and often requires immunohistochemical staining for differentiation.In recent years,convolutional neural networks have gradually emerged,and in the medical field,deep learning models based on various image resources such as CT images,MRI images,ultrasound,and pathology have been applied in clinical settings using convolutional neural networks.Objective:To collect clinical information and pathological data,and verify the correlation between lymphovascular invasion and prognosis and lymph node metastasis of hypopharyngeal squamous cell carcinoma through statistical analysis;to construct a deep learning model based on H&E staining to achieve automatic classification of lymphovascular invasion of hypopharyngeal squamous cell carcinoma and apply it in clinical practice.Methods:Clinical and pathological data of patients diagnosed with hypopharyngeal squamous cell carcinoma and treated with curative surgery in the Department of Otolaryngology Head and Neck Surgery,Qilu Hospital of Shandong University from January 2016 to June 2020 were collected,screened according to inclusion and exclusion criteria,and the data of selected patients were obtained,including age at onset,gender,smoking and alcohol history,tumor location,degree of differentiation,maximum tumor diameter,T staging,lymph node metastasis,and vascular invasion.The patients were followed up by phone to obtain their survival status and survival time.The relationship between lymphovascular invasion and prognosis was analyzed using a proportional hazards regression model.The relationship between lymphovascular invasion and lymph node metastasis was analyzed using a generalized linear regression model.Pathological specimens corresponding to patients were obtained from the pathology department specimen bank of Qilu Hospital and digitized.The lymphovascular invasion,and non-tumor-infiltrated vessels and lymphatics on the digitized pathological slices were labeled.The digitized slice data were preprocessed,segmented into patches,and divided into training set,validation set,and test set according to 70%,15%,and 15%,respectively.Convolutional neural networks were used for training,and the model performance was evaluated using the ROC curve and confusion matrix.Results:A total of 117 cases were included in the study.Cox regression analysis identified that vascular invasion(HR=2.133,95%CI:1.105-4.118,P<0.05)was an independent risk factor affecting patient survival,and patients with positive lymphatic vessel invasion had a worse prognosis.In addition,tumor maximum diameter was also an independent risk factor affecting patient prognosis.Logistic regression analysis showed that vascular invasion(OR=12.917,95%CI:2.544-236.099,P<0.05)was an independent risk factor for lymph node metastasis,and differentiation degree and tumor location were also associated with lymph node metastasis.The deep learning model had a good classifier performance with an area under the ROC curve of 0.978,sensitivity of 0.875,specificity of 0.993,positive predictive value of 0.983,negative predictive value of 0.944,and accuracy of 0.955 in the test set.The confusion matrix showed an error rate of nearly 1%in patches without tumor vascular invasion,and an error rate of 12%and a recall rate of 88%in patches with tumor vascular invasion.Conclusion:Vascular invasion is a significant predictor of poor prognosis in patients with hypopharyngeal squamous cell carcinoma,and is closely associated with lymph node metastasis.The deep learning model based on H&E staining can accurately classify vascular invasion in hypopharyngeal squamous cell carcinoma,providing a potential tool for clinical decision-making and improving patient outcomes.The study validated that lymphovascular invasion is an independent risk factor for the prognosis of patients with hypopharyngeal squamous cell carcinoma,which is closely related to lymph node metastasis of hypopharyngeal squamous cell carcinoma.The artificial intelligence model established by deep learning for automatic identification and classification of lymphovascular invasion in hypopharyngeal squamous cell carcinoma demonstrates excellent classification performance.