Research On Prognostic Model Analysis And Treatment Plan Recommendation Of Cervical Cancer Based On Deep Learning

Part Ⅰ Escalated radiation and prophylacticextended field nodal irradiation arebeneficial for FIGO ⅢB cervical cancer patients’ prognosisBackground:Currently,the standard treatment for locally advanced cervical cancer patients is concurrent chemoradiotherapy.Here we aim to evaluate therapeutic efficacy,treatment failure,toxicity and prognostic factors for FIGO ⅢB cervical cancer patients.Methods:A comprehensive retrospective analysis was performed to understand various factors which contribute to ⅢB cervical cancer prognosis.In total 223 well defined patients were assigned according to their pathological subtype,age,pre-treatment HGB level,tumor size,pelvic lymph node(LN)metastasis,para-aortic LN metastasis as well as external irradiation technologies,treatment duration,point A EQD2 dose and concurrent chemotherapy cycles.We then performed correlation studies of these factors and OS,DFS,LCR,DMFS using univariate and multivariate analysis respectively.Results:We managed to achieve 207(92.8%)complete response(CR)and 16(7.2%)partial response(PR)with acceptable adverse effects.Notably,the 5 years OS,DFS,LCR,DMFS for these patients were 61.1,55.2,83.6 and 66,4%respectively.Importantly,our studies suggest that escalated point AEQD2 can significantly improve OS,DFS and LCR for FIGO ⅢB cervical cancer patients,furthermore,patients without para-aortic LN metastasis who received prophylactic extended field irradiation have significant survival advantage for DFS and a tendency to improve OS and DMFS.Conclusions:Our results suggest that FIGO ⅢB cervical cancer patients should receive higher EQD2(≥98Gy10)radiotherapy,moreover,patients without para-aortic LN metastasis should receive prophylactic extended field nodal irradiation to improve prognosis.Part Ⅱ Evaluation of the efficacy of prophylactic extended field irradiation in the concomitant chemoradiotherapy treatment of locally advanced cervical cancer,stage ⅢB in the 2018 FIGO classificationBackground:The new staging system of cervical cancer issued in 2018 by the International Federation of Gynecology and Obstetrics(FIGO),calls for a new evaluation of the efficacy of prophylactic extended field irradiation(EFI)in the concomitant chemoradiotherapy/brachytherapy treatment of locally advanced cervical cancer patients(stage ⅢB).Methods:We performed a retrospective study consisting of 133 FIGO ⅢB cervical cancer patients treated in the Peking Union Medical College Hospital from 2002 to 2010.The patients were distributed in two groups depending whether they were treated with EFI or pelvic only irradiation.The therapeutic efficacy,toxicity and prognostic factors of EFI were evaluated in the frame of the new FIGO staging system.Results:When compared to patients who received pelvic only irradiation,patients who received prophylactic EFI showed significantly less distant metastasis and a significant improvement in their 5 years overall survival(OS),disease free survival(DFS),out of field recurrence free survival(OFRFS)and para-aortic lymph node metastasis free survival(PALNMFS).Multivariate analysis determined that EFI is an independent prognosis factor for DFS,OFRFS and PALNMFS.Finally,although more acute complications were observed in the EFI group,there is no significantly worst acute toxicity in the EFI group.Conclusion:Our retrospective analysis supports the prophylactic effect of EFI in the concomitant chemoradiotherapy treatment of ⅢB patients and suggests that this prophylactic effect is associated with a clear improvement in 5-years OS,DFS,OFRFS and PALNMFS.Consequently,EFI appears to be a very valid treatment option for FIGO ⅢB cervical cancer patients.Part Ⅲ Machine learning to predict local recurrence and distance metastasis of locally advanced cervical cancer after radiotheraphy alone or concurrent chemoradiotherapyBackground:Radiotherapy is a widely recommended treatment for locally advanced cervical cancer,which is the most common type of cervical cancer in China and other developing countries.The main cause of treatment failure is known to be local recurrence and distance metastasis.This study aimed to develop a predictive model for local recurrence and distance metastasis after definitive radiotheraphy or concurrent chemoradiotherapy,which could play a key role in treatment plans making.Methods:1421 consecutive locally advanced cervical cancerpatients who received definitive radiotherapy of concurrent chemoradiotherapy in Peking Union Medical College Hospital were analyzed.Using clinical characteristics and postoperative histopathology of cervical cancer patients,models were generated using machine learning methods of random forest,adaboost and logistic regression.Finally,a combined(stacking)model of these was generated.The relative importance of factors to outcome was calculated as a percentage contribution to the model.Results:All the models show good discrimination in predicting OS,DFS,LC and DMFS with time over 5 years.Out of the four predictive models,the accuracies for all metrics were similar,with the stacking model performing the best(82.67%for OS,80.7%for DFS,87.35%for LC,88.06%for DMFS).Performance was also similar when evaluated under the receiver operating characteristic(ROC)curve(AUCs).In the final model,the most important variables were tumor stage and para-aortic lymph node metastasis for DFS and LC,chemotherapy times for OS and the SCC for DMFS.Conclusion:The machine learning models trained with a handcraft cervical cancer dataset provide excellent performance in predicting the status of OS,DFS,LC and DMFS,and will be valuable in clinic.Part Ⅳ Study on the establishment of cervical cancer radiotherapy prognosis prediction model and clinical individualized treatment plan recommendation based on deepreinforcement learning and structured clinical dataCervical cancer is one of the most common gynecological malignancies.While in developed countries patients are usually diagnosed at an early stage,patients in our country are usually at an advanced stage when they are diagnosed.The main treatment for the advanced cervical cancer is radiotherapy and chemotherapy.Although the application of concurrent chemoradiation has significantly improved the prognosis of patients with cervical cancer of advanced stages,20-30%of the patients still suffer from recurrence or metastasis within 2 years after treatment,which seriously affects the further improvement of the prognosis of cervical cancer.This is a worldwide difficult and hot issue.There has been no significant breakthrough so far in accurately predicting the prognosis of patients based on their characteristic performance and indicators,and choosing the most appropriate treatment method accordingly.In recent years,a new field of artificial intelligence known as deep reinforcement learning has emerged,which can build analysis and learning neural networks that simulate the human brain.Deep reinforcement learning brings better accuracy of processing large amounts of data compared with traditional methods,and it has shown outstanding disease prognosis prediction capabilities in other fields.However,it has not yet been reported on whether it can accurately predict the recurrence and metastasis of cervical cancer after radiotherapy,provide accurate survival prediction,and formulate a personalized treatment plan based on the actual situation of the patient.This research is based on the cervical cancer diagnosis and treatment data of Peking Union Medical College Hospital for many years and is to explore the development of a high-efficiency treatment plan recommendation technology based on feature selection,a personalized treatment plan recommendation technology based on reinforcement learning,and the efficient integration of the two technologies in order to improve curative and prognosis.