RESUMO
This study aimed to develop a robust multiclassification pipeline to determine the primary tumor location in patients with head and neck carcinoma of unknown primary using radiomics and machine learning techniques. The dataset included 400 head and neck cancer patients with primary tumor in oropharynx, OPC (n = 162), nasopharynx, NPC (n = 137), oral cavity, OC (n = 63), larynx and hypopharynx, HL (n = 38). Two radiomic-based multiclassification pipelines (P1 and P2) were developed. P1 consisted in a direct identification of the primary sites, whereas P2 was based on a two-step approach: in the first step, the number of classes was reduced by merging the two minority classes which were reclassified in the second step. Diverse correlation thresholds (0.75, 0.80, 0.85), feature selection methods (sequential forwards/backwards selection, sequential floating forward selection, neighborhood component analysis and minimum redundancy maximum relevance), and classification models (neural network, decision tree, naïve Bayes, bagged trees and support vector machine) were assessed. P2 outperformed P1, with the best results obtained with the support vector machine classifier including radiomic and clinical features (accuracies of 75.3 % (HL), 75.4 % (OC), 71.3 % (OPC), 92.9 % (NPC)). These results indicate that the two-step multiclassification pipeline integrating radiomics and clinical information is a promising approach to predict the tumor site of unknown primary.
RESUMO
BACKGROUND: Pituitary adenomas and craniopharyngiomas are the most common lesions of the sellar region. These tumors are responsible for invasion or compression of crucial neurovascular structures. The involvement of the pituitary stalk warrants high rates of both pre- and post- operative diabetes insipidus. The aim of our study was to assess the accuracy of machine learning analysis from preoperative MRI of pituitary adenomas and craniopharyngiomas for the prediction of DI occurrence. METHODS: All patients underwent MRI exams either on a 1.5- or 3-T MR scanner from two Institutions, including coronal T2-weighted (T2-w) and contrast-enhanced T1-weighted (CE T1-w) Turbo Spin Echo sequences. Feature selection was carried out as a multi-step process, with a threshold of 0.75 to identify robust features. Further feature selection steps included filtering based on feature variance (threshold >0.01) and pairwise correlation (threshold <0.80). A Bayesian Network model was trained with 10-fold cross validation employing SMOTE to balance classes exclusively within the training folds. RESULTS: Thirty patients were included in this study. In total 2394 features were extracted and 1791 (75%) resulted stable after ICC analysis. The number of variant features was 1351 and of non-colinear features was 125. Finally, 10 features were selected by oneR ranking. The Bayesian Network model showed an accuracy of 63% with a precision of 77% for DI prediction (0.68 area under the precision-recall curve). CONCLUSIONS: We assessed the accuracy of machine learning analysis of texture-derived parameters from preoperative MRI of pituitary adenomas and craniopharyngiomas for the prediction of DI occurrence.