Automatic Diagnosis And Localization Of Cervical Cancer On Medical Images

Cervical cancer is a common gynecological tumor disease,which seriously threatens women’s life and health.Since cervical cancer can be cured,early inspection and treatment can improve the survival rate of patients.At present,cervical cancer is diagnosed mainly by pathological examination,and doctors then delineate tumor lesions and metastatic lymph node lesions based on abdominal CT(computed tomography)images for surgery or radiation therapy.Since pathological examination will bring physical and mental damage to patients,and cervical cancer has obvious characteristics on CT images,we consider only use CT images to achieve cervical cancer diagnosis and focus location.At present,the localization of lesions on CT images mainly relies on manual reading,which not only requires doctors to have rich theory and experience,but also causes work burdens on doctors,and there are inevitably subjective differences.In this study,we apply computer-aided diagnosis to cervical cancer diagnosis and lesion location.This task is briefly divided into cervical detection and lymphatic detection,which can not only diagnose the disease non-invasively,but also quickly locate the lesion area.On CT images,there are obvious differences in characteristics between normal and diseased cervix,but the size and shape of metastatic lymph nodes are variable and there are many interference items.These non-obvious features are the difficulty in the task of lymph node detection.Based on the classic target detection network Faster R-CNN,this paper proposes a method for the automatic diagnosis and lesion location of cervical cancer,which provides accurate and objective references for the diagnosis and treatment of clinical doctors.In this Study,we simulate the clinical process of cervical cancer diagnosis and lesion location,and divide the project into two parts: cervical detection and lymphatic detection.In cervical detection task,we will directly use the target detection network Faster R-CNN,and cervical cancer diagnosis is based on the threshold of the number of sheets judged by the category.The experimental results prove the accuracy of the diagnosis method.In lymphatic detection task,the characteristics of metastatic lymph nodes in CT images are first analyzed,and a series of improvements have been made to Faster R-CNN in a targeted manner,including the preprocessing of data images,the introduction of feature pyramids and attention mechanisms,and the introduction of three-dimensional information between sequences and post-process assist in removing false positives.The comparative experiments show that the improved network in this paper can accurately locate lymphatic lesions,and the final accuracy rate reaches about98%.Compared with the results of manual reading by doctors,the proposed lymphatic detection model has certain advantages in detection precision and recall,and it takes less time,which reflects its effectiveness in clinical adjuvant treatment.The automatic diagnosis and lesion location method for cervical cancer proposed in this article can not only accurately diagnose cervical cancer based on CT images and reduce the patient’s pain,but also accurately locate the lesion area,providing an objective reference for clinical treatment.It can reduce the burden on doctors and play a guiding and teaching role for inexperienced doctors.At the same time,clinicians have confirmed the effectiveness of our proposed method.After further improving the accuracy of the detection according to clinical requirements,it is expected to be directly applied to clinical treatment guidance.And the proposed detection method for multimorphological targets can also be extended to the detection of other organs or tissues,and its potential can be further explored.