Total mesorectal excision(TME)following neoadjuvant therapy is the standard treatment for patients with locally advanced rectal cancer(LARC).After neoadjuvant therapy,approximately 50% to 60% of patients with LARC achieve tumor stage reduction and approximately 20% achieve pathological complete response.Subsequent TME is effective at providing local tumor control.However,there are still some challenges in the implementation of this treatment.The first challenge is how to select patients suitable for neoadjuvant therapy.Studies have shown that approximately 7% of patients with LARC do not respond to neoadjuvant therapy.If these non-responders can be identified before neoadjuvant therapy,they will have chance to avoid this ineffective and painful overtreatment and to undergo surgery earlier.Whether all T3 rectal cancer patients need neoadjuvant therapy remains controversial.The European Society for Medical Oncology guidelines recommend lymph node status as one of the determinants of neoadjuvant therapy.Therefore,accurate assessment of lymph node status is of great significance for the selection of treatment schedule for patients with T3 rectal cancer.The second challenge is how to determine the extent of surgical resection after neoadjuvant therapy.TME is an unnecessary overtreatment for patients with good therapeutic response,and local excision may be a more appropriate surgical procedure for these good responders.One of the conditions for local excision is that the regional lymph nodes around the tumor must be pathologically negative.Therefore,accurate assessment of the regional lymph node status after neoadjuvant therapy is crucial for determining the extent of surgical resection.In this paper,based on the imaging data in diagnosis and treatment,specific radiomics-based algorithms were developed to assist the decision of treatment plan for patients with LARC.The main contributions of the present study are as follows:First,a radiomics-based algorithm for pre-therapeutic prediction of non-response to neoadjuvant therapy in locally advanced rectal cancer was developed.Specifically,we retrospectively enrolled 425 patients with LARC who received neoadjuvant therapy before TME.All patients underwent multi-sequence magnetic resonance imaging(MRI)scanning before neoadjuvant therapy.Then patients were randomly allocated into a training set and a validation set.High-dimensional radiomic features were extracted from each imaging sequence.Then,a multi-sequence radiomics model was established in the training set through a three-step feature selection procedure,which consists of successively conducting Wilcoxon rank-sum test,Spearman correlation analysis,and least absolute shrinkage and selection operator(LASSO)-based logistics regression.Three single-sequence radiomics models were established according to the same procedure as above.The results of receiver operating characteristic(ROC)curve analysis in validation set demonstrated that the multisequence-based modeling method is superior to the single-sequence-based modeling method.The developed multi-sequence-based model achieved an area under the curve(AUC)value of 0.773.Finally,pre-therapeutic MRIs of 125 breast cancer patients who received neoadjuvant therapy and radical surgery from a single hospital were retrospectively collected,and the same modeling and validation procedures as used in the LARC dataset were performed to demonstrate the generalizability of the non-response prediction algorithm framework proposed in this study.Secondly,a radiopathomics-based algorithm for predicting the lymph node status in T3 rectal cancer was developed.In this study,a modeling method for integrating pre-therapeutic computed tomography(CT)imaging information and biopsy digital pathological imaging information was proposed.To be specific,this study first retrospectively collected initial diagnostic CT images,biopsy digital pathological images and clinical information of 85 patients with T3 rectal cancer who had undergone TME surgery and lymph node dissection from a single hospital.Then patients were randomly allocated into a training set and a validation set.Subsequently,high-dimensional radiomic features of the intratumoral volumes and peritumoral volumes were extracted from CT images,and high-dimensional pathomic features were extracted from the biopsy digital pathological images.Then,in the training set,the key radiopathomic features were selected by LASSO.A radiopathomics model was constructed by support vector machine using the selected key radiopathomic features to evaluate the risk of lymph node positivity.Then a radiomics model and a pathomics model were established respectively through the same procedure as above.Finally,ROC curve analysis of the three models in the validation set showed that this radiopathomics modeling method is superior to radiomics or pathomics modeling method,and the developed radiopathomics model can greatly help to improve the diagnostic accuracy of radiologist from 60% to 80%.Finally,a radiomics-based algorithm was developed to predict the lymph node status following neoadjuvant therapy for patients with LARC.In this study,a radiomics modeling method based on multi-sequence MRI scanned after neoadjuvant therapy and before TME was proposed.To be specific,this study first retrospectively collected post-therapeutic multisequence MRI images of 391 patients with LARC who received neoadjuvant therapy and TME and underwent lymph node dissection.These patients patients were randomly allocated into a training set and a validation set.Then high-dimensional radiomic features of the tumor region were extracted from multi-sequence MRI images.A multi-sequence radiomics model was established in the training set through a three-step feature selection procedure,which consists of successively conducting Wilcoxon rank-sum test,Spearman correlation analysis,and LASSO-based logistics regression.Then the diagnostic performance of the radiologists and the radiomics model was compared in the validation set.Finally,a combined model was established by incorporating radiologist's diagnostic results and radiomic signature.Stratified analysis showed that the combined model established in this study achieved a very good predictive ability in post-therapy MRI T1-2 patients with an AUC value of 0.957 and a negative predictive value of 100%.To sum up,in view of the challenges during the decision of treatment plan for LARC,the present study developed radiomics-based decision algorithms to provide valuable information for the personalized treatment plan. |