Radiomics Predicting Lymph Node In Operable Cervical Cancer

Background and PurposeCervical cancer has been ranked as the fourth most common cancer in women worldwide,where 85%of these patients occur in developing countries.In these developing countries,cervical cancer is the leading cause of cancer related deaths.It is important to clarify the lymph node status preoperatively in cervical cancer,which could facilitate the treatment planning and prognosis making.Preoperative imaging tests including CT scan and MRI exams may have some values in the prediction of lymph node metastases.Radiomics have drawn much more attentions in recent years and it is a process of converting medical images into mineable high dimensional data in a high throughput manner which could be further analyzed for clinical decision making.Previous studies have shown that these objective and quantitative radiomic features could be used as meaningful biomarkers for the prediction of treatment response and prognosis in various types of cancer including the cervical cancer.However,the main limitation of popular radiomic studies is that the performance of these predictive model almost always could not reach the clinical useful level since they are often based on the conventional machine learning method and limited numbers of variables.Deep learning(DL)is a subspecialty of the machine learning and artificial intelligence,which has shown impressive performance in medical diagnosis,treatment response prediction,and prognosis making.Yet,it is unclear how we could construct the optimal DL model in the radiomic studies,especially for the lymph node metastases prediction in cervical cancer.Thus,we here explored the possibility of developing a deep learning radiomic model which could incorporate preoperative radiomic and clinicopathological features.The established model may be useful in identify those patients who would be spared for the unnecessary hysterectomy surgery in future.MethodsThis research collected and retrospectively analyzed a cohort of 226 pathological proven operable cervical cancer patients in two academic medical institutions from December 2014 to November 2017.All patients were diagnosed as stage IA to stageⅡB cervical cancer confirmed by histopathology according to the FIGO staging system 2019,and received surgical resection without prior therapy.Then this dataset was split into training set(n=176)and independent testing set(n=50)randomly.Five radiomic features were selected and a radiomic signature was established.We then combined these five radiomic features with the preoperative tumor histology and grade of these patients together.Baseline logistic regression model(LRM)and support vector machine model(SVM)were established for the comparison.We then explored the performance of a deep neural network(DNN),which is a popular deep learning model nowadays.Finally,performance of this DNN was validated in another independent test set including 50 cases of operable cervical cancer patients.ResultsOne thousand forty-five radiomic features were extracted for each patient,which covered 5 aspects of radiomics including the intensity,shape,texture,wavelet and log transformation..Twenty-eight features were found to be significantly correlated with the lymph node status in these patients(P<0.05).Five of these 28 radiomic features were selected for further study due to their higher value of C index.A radiomic signature was established based on these 5 features which had area under receiver operating characteristic curve(ROC)of 0.7154 and accuracy of 88.07%.Baseline logistic regression model included these radiomic features as well as histology and grade,which had area under ROC of 0.7372 and accuracy of 89.20%.Then,the above 7 features were included into the DNN model.This DNN model had 4 neural layers,in which each layer had 10 neurons.For this DNN,we used Adagrad optimizer and had 1500 iterations for optimal performance.We achieved area under ROC of 0.99 and accuracy of 97.16%in the internal validation.To exclude inevitable overfitting,we also performed the independently external validation.Area under ROC of 0.90 and accuracy of 92.00%was obtained in the 50 cases of test set.ConclusionsThis study used the deep learning model to provide a comprehensive predictive model including radiomic features,histology and grade.This model showed an acceptable accuracy in predicting the status of lymph node in operable cervical cancer patients.