CNN-based computer vision architectures for anomaly visualization in medical images: A technical review
Abstract Early disease detection, precise prognosis, and effective treatment planning heavily depend on medical image diagnosis. Because manual image analysis requires substantial clinical expertise, computer-aided systems have become essential. This study examines various Convolutional Neural Network (CNN) algorithms for anomaly detection and classification in medical scans. Following a systematic search across the IEEE Xplore, Web of Science, and Google Scholar databases, 91 publications were included in our synthesis. We provide a structured comparative analysis of prominent networks, including U-Net, ResNet, GoogLeNet, and customized models, evaluating their applications, developments, and implementation challenges. The findings of this review demonstrate that fine-tuning key architectural parameters can significantly improve core diagnostic metrics, specifically accuracy, sensitivity, F1 score, precision, Dice Score Coefficient, and AUC . Ultimately, this work maps the current state of deep learning in healthcare and outlines critical areas for future inquiry.
Authors
- Ruban Nersisson (ORCID: https://orcid.org/0000-0003-1695-3618)
- Iyyappan Sagadevan (ORCID: https://orcid.org/0009-0000-9469-2914)
Institutions
- Vellore Institute of Technology University (IN)
Publication Details
- Journal
- Imaging
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1556/1647.2026.00428
- Primary Topic
- Medical Image Segmentation Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00