The Application of Transfer Learning in Medical Image Classification: A Review of Literature

Abstract

Timely identification about health disorders has been made possible by photographic technologies, therefore photographic evidence provides a helpful technique towards illness identification. In addition to being labor-intensive, mechanical imagery processing techniques are prone to inter- and intra-observer inconsistency. Those restrictions may be addressed by computerized diagnostic imagery investigation methods. Transfer Learning (TL) frameworks using computerized health-related imagery processing have been looked at within the paper. It is has found how transfer learning can be used for many different healthcare scanning responsibilities, including recognizing objects, diseases classification, separation, and sensitivity scoring, etc. In contrast to conventional deep learning techniques, it is demonstrated that transfer learning offers superior selection assistance and uses reduced experimental input. A total of 100 peer-approved English language publications using the archives, IEEE Xplore and PubMed, where obtained till April 2025. A total of 53 research papers are considered suitable to the subject matter of this study after the PRISMA procedures for article screening was applied. Transfer learning techniques, such as pattern extraction, pattern extraction mix, fine tuning, and fine tuning from the start, are examined by works that concentrated on choosing core algorithms. Most research conducted experimental evaluations about several algorithms, then examined shallower depth approaches. Depth framework that has been used frequently within the academic research is Inception. In order to determine the best arrangement for the Transfer Learning, most research quantitatively compared several methods. just one methodology was used throughout the remaining experiments, then the two highly popular methods included pattern extraction and fine-tuning from start. Some research used pretrained algorithms for fine tuning and extraction of features hybrids. According to this study, the best popular Transfer Learning algorithms in analyzing medical images are AlexNet, ResNet, VGGNet, and GoogleNet. Such TL algorithms have been shown to be capable of comprehending healthcare photos, plus their capability to do so is improved by customisation, rendering them valuable instruments in studying photo scans.