Recent Advances in DeepLabV3-DenseNet Leveraging Radiomics Feature Extraction and Non-Invasive Detection of Microsatellite Instability in Colorectal Cancer with a Hyperparameters-Tuned Pre-trained Model: A Systematic Review

Abstract

Microsatellite instability (MSI) is a critical biomarker in colorectal cancer (CRC), influencing prognosis and therapeutic decision-making, particularly for immunotherapy. Traditional MSI detection methods rely on invasive tissue biopsies and molecular testing, which are time-consuming and costly. Recent advances in artificial intelligence (AI), deep learning, and radiomics have enabled non-invasive MSI prediction using medical imaging and histopathological data. This systematic review explores recent developments in MSI detection using deep learning architectures such as DeepLabV3, dense Net, and hybrid frameworks combined with radiomics feature extraction. Studies demonstrate that convolutional neural networks (CNNs) and transformer-based models can accurately predict MSI status directly from histopathological slides and imaging modalities. For example, deep learning systems have achieved high diagnostic performance with AUROC values exceeding 0.95 in large-scale validation cohorts. Additionally, radiomics-based approaches using CT and PET imaging have shown promising results in non-invasive MSI prediction by extracting high-dimensional quantitative features. Hybrid models integrating feature extraction, segmentation (DeepLabV3), and classification (dense Net) further enhance predictive accuracy by capturing spatial and contextual information. Despite these advancements, challenges such as data heterogeneity, computational complexity, and clinical validation remain. This review provides a comprehensive analysis of methodologies, comparative performance, and future directions for AI-driven MSI detection.