Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2026
|
| Online Access: | https://doi.org/10.5281/zenodo.19635204 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p class="MsoNormal"><span>The use of brain magnetic resonance imaging (MRI) for detecting anomalies is highly significant for early diagnosis of neurological disorders, including tumors, lesions, and degenerative diseases. However, supervised deep learning models are sensitive to large amounts of annotated data, which are costly and time-consuming to obtain in the medical sector. To avoid this limitation, the paper will discuss how to apply self-supervised learning (SSL) to detect anomalies in brain MRI scans. The proposed approach will work with raw data to learn powerful representations of normal brain anatomy using techniques such as masked image reconstruction and contrastive learning. It identifies abnormalities by modelling the distribution of normal scans based on aberrations in reconstruction error or feature space. To enhance the quality of representation and generalization for MRI modalities, this model integrates a contrastive goal-based encoder-decoder design, employing a hybrid architecture. Experimental studies show that models obtained with SSL outperform traditional unsupervised and supervised baselines in detecting subtle and hidden anomalies. The findings reiterate that the potential of SSL lies in reducing dependence on annotated data and improving detection performance, thereby enabling scalable, clinically-relevant diagnostic systems.</span></p>