A Comprehensive Review of DeeplabV3-DenseNet: Leveraging Radiomics Feature Extraction and Non-Invasive Detection of Microsatellite Instability in Colorectal Cancer with a Hyperparameters-Tuned Pre-trained Model

Abstract

Microsatellite instability (MSI) is a critical biomarker in colorectal cancer (CRC) that plays a significant role in prognosis, treatment selection, and immunotherapy response. Traditional MSI detection methods rely on invasive tissue sampling and molecular testing, which can be time-consuming and costly. Recent advancements in artificial intelligence, particularly deep learning and radiomics, have enabled non-invasive MSI prediction using medical imaging data such as CT, MRI, and histopathological images. Radiomics extracts high-dimensional quantitative features from imaging data, capturing tumour heterogeneity and microenvironment characteristics. Studies have shown that radiomics-based models achieve strong predictive performance with area under the curve (AUC) values ranging from 0.78 to 0.96 in MSI detection. Deep learning architectures such as Dense Net and DeeplabV3 have further enhanced feature extraction and segmentation capabilities. DeeplabV3 enables precise tumour segmentation through atrous convolution, while Dense Net facilitates efficient feature reuse and gradient propagation. The integration of these models with hyperparameter tuning and transfer learning improves prediction accuracy and generalization. This paper presents a comprehensive review of DeeplabV3-DenseNet-based frameworks combined with radiomics for non-invasive MSI detection in colorectal cancer. It highlights recent trends, methodological advancements, challenges, and future research directions in AI-driven radio genomics for precision oncology.