Research On Automatic TNM Staging Technology Of Chinese Pathology Reports

Cancer staging is an important link in cancer diagnosis and treatment,and it is an essential basis for clinicians to formulate treatment plans and evaluate prognosis.Cancer staging can be divided into clinical staging and pathological staging,and clinicians rely on the cancer staging system for staging.There are different types of cancer staging systems,among which TNM(Tumor Node Metastasis)staging system is the most common and widely used for malignant tumors,T,N,and M represent the status of the primary tumor,the involvement of regional lymph nodes,and the status of distant metastasis,respectively.In recent years,scholars at home and abroad have carried out a series of studies on TNM staging for pathology reports.At present,the current domestic researches on TNM staging for pathology reports have the following problems: 1)Most studies are aimed at clinical staging;2)Insufficient explanability of staging results.In addition,clinicians are easy to make mistakes when using the electronic medical record system for TNM staging,at the same time,it is difficult to carry out batch processing due to the time-consuming and labor-consuming manual staging.Regarding the issues above,this article proposes an automatic TNM staging technology for Chinese pathology reports.The technology includes two steps: first,named entity recognition(NER)for pathology reports,and then on the basis of named entity recognition,TNM staging combining information extraction and rules.This article uses postoperative pathology report data of lung cancer patients to construct a corpus and conducts research on Chinese pathology report automatic TNM staging technology,and finally developes the TNM automatic staging system for Chinese pathology reports.This paper has completed the following main tasks:(1)This paper proposes a named entity recognition(NER)algorithm for pathology reports based on improved Flat-Lattice Transformer(FLAT)model.The FLAT model improves the Transformer model so that it can accept the Lattice structure as input,add lexical information,and use a new relative position code to replace the original code,so that the model can perceive distance and direction.In addition,we learn from Revnet to reduce the memory consumption of the model.According to the lung cancer staging guidelines,the annotation specification of Chinese pathology reports was formulated,and194 pathology reports were annotated to construct the corpus of this paper.The self-built corpus was used to design comparative experiments to verify the performance of the model.Finally,the F1 value of the improved FLAT model in the entity recognition task reached 97.19%,showing a better recognition performance than the unimproved FLAT model and the comparison models(BILSTM,Lattice LSTM and TENER).(2)This paper proposes a TNM staging algorithm combining information extraction and rules.On the basis of named entity recognition results,it carries out negative information recognition and reasoning based on heuristic rules to obtain staging evidences and use them for staging.This article uses 100 manually labeled pathology reports to summarize the rules and verify the performance of the staging algorithm.In the end,the accuracy and recall F1 values of the automatic staging algorithm in this article are 93.27%,91.85%,and 92.55%,respectively,which proves the effectiveness of our algorithm.(3)This paper designs and developes an automatic staging system for Chinese pathology reports.The system has realized the functions of managing pathology reports,staging pathologyl reports automatically and the function of visually displaying staging evidences.It can not only assist doctors in diagnosis,but also can be used for medical quality control.