ADVANCEMENTS IN TRANSFER LEARNING: A COMPREHENSIVE REVIEW OF NOVEL APPROACHES FOR MRI BRAIN IMAGE DIAGNOSIS

• Published in: Science journal of University of Zakho (Online) Vol. 13; no. 3; pp. 334 - 342
• Main Authors: Waysi, Diyar, Ahmed, Berivan Tahir, Yasin, Hajar Maseeh
• Format: Journal Article
• Language: English
• Published: University of Zakho 02.07.2025
• Subjects: Deep learning; Diagnose Diseases; machine learning; MRI Image; Training Algorithm; Transfer Learning
• ISSN: 2663-628X; 2663-6298; 2663-6298
• DOI: 10.25271/sjuoz.2025.13.3.1470

Magnetic Resonance Imaging (MRI) has rapidly advanced and established itself as an indispensable tool in both the detection and diagnosis of several diseases, most notably brain tumors. The interpretation of MRI scans still largely relies on expert radiologists, which can be time-consuming and potentially subject to variability. Transfer learning (henceforth, TL) approaches show potential for improving diagnostic precision in medical imaging analysis. In this literature review, the potential of MRI scans in classifying and detecting various medical conditions, such as glioma and Alzheimer’s, is discussed alongside current algorithmic limitations. Current research indicates potential challenges in adapting existing supervised deep learning algorithms that process MRI images to more efficient approaches. The findings suggest a notable increase in the quality of detecting sub-pathologies, even with a scarcity of well-annotated images. This can potentially reduce the training cycle duration. When transfer learning is applied to diagnostic approaches, it may act as supplemental support for decision-making processes for tumorous growth detection, potentially reducing the time period for treatment and increasing effectiveness according to preliminary research. This review examines the expansion in transfer learning in MRI for the assessment and treatment of brain disorders through recent algorithms from the current literature.