Document Type : Review
Authors
1
Department of Medical Informatics, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran.
2
Artificial Intelligence in Medical Sciences Research Center, Smart University of Medical Sciences, Tehran, Iran
3
Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
4
Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran.
5
Department of Medical Informatics, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
6
Research Center for Nuclear Medicine, Tehran University of Medical Sciences, Tehran, Iran.
10.30476/jhmi.2026.110527.1354
Abstract
Background:
The global rise in cancer incidence highlights the urgent need for more effective diagnostic tools to support early detection and optimize treatment planning. Deep learning-based object detection (OD) has emerged as a significant advancement in medical imaging, automating the identification and classification of lesions. By improving the analysis of CT, MRI, and PET/CT scans, OD reduces interpretation time, enhances diagnostic accuracy, and promotes more consistent clinical decision-making in oncology.
Objective:
This review maps the clinical applications of deep learning-based OD in cancer imaging, evaluates the diagnostic performance of various architectures, and identifies barriers to routine clinical adoption.
Methods:
Following Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines, PubMed and Scopus were systematically searched for studies published between 2014 and January 2026. Peer-reviewed studies applying deep learning-based OD models to cancer imaging using CT, MRI, or PET/CT were included. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool.
Results:
Of the 46 included studies, 27 utilized one-stage detectors, 11 used two-stage detectors and 5 employed both approaches. One-stage models were favored for their efficiency, while two-stage models demonstrated superior detection of small or low-contrast lesions. Applications included early cancer detection, lesion classification, and evaluation of treatment response.
Conclusions:
Deep learning-based OD demonstrates high diagnostic accuracy and potential to optimize oncologic imaging workflows. However, challenges such as validation, generalizability, and integration into clinical practice limit widespread adoption. Addressing these barriers is critical to realizing the full potential of OD in improving cancer management.
Keywords
Main Subjects
)