Identifying Optimal Multiparametric MRI Sequence Combinations for Prostate Cancer Classification: An Integrated Deep Feature and Machine Learning Approach

Prostate cancer (PCa) is a leading cause of cancer-associated deaths among men worldwide. Accurate classification of PCa into clinically significant (CS) and clinically insignificant (CiS) is crucial for guiding treatment strategies. Multiparametric magnetic resonance imaging (mpMRI) has become a valuable tool for detecting and characterizing PCa lesions. This study proposes a novel framework to identify the optimal combinations of mpMRI sequences for PCa classification in the entire prostate and specific anatomical zones using transfer learning with pre-trained VGG19 and Vision Transformer (ViT) models for feature extraction. The framework exhaustively evaluates all possible combinations of features extracted from nine mpMRI sequences. An SVM classifier is employed for classification, and the best-performing combinations are selected based on evaluation metrics. Additionally, it investigates the importance of each mpMRI sequence by analyzing its frequency in the top-performing combinations. The proposed framework achieves promising results, with the best-performing combination yielding an AUC score of 0.87 for the entire prostate. The results highlight the significance of high b-value DWI sequences, particularly BVAL (b=1400 s/mm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), followed by Ktrans, in contributing to PCa classification performance. The identified optimal combinations and sequence importance analysis provide valuable insights for PCa diagnosis using mpMRI.