Knee Injury Detection Using Deep Learning on MRI Studies: A Systematic Review.
Athanasios Siouras1,2, Serafeim Moustakidis3, Archontis Giannakidis4
1Department of Computer Science and Biomedical Informatics, School of Science, University of Thessaly, 35131 Lamia, Greece.
Diagnostics (Basel, Switzerland)
|February 25, 2022
Summary
Deep learning models show high accuracy (72.5-100%) for detecting knee injuries like ACL tears and meniscus damage on MRI scans. Further research is needed to address limitations for widespread clinical use.
Area of Science:
- Orthopedics
- Radiology
- Artificial Intelligence
Background:
- Accurate and cost-effective detection of knee injuries is crucial for improved treatment.
- Deep learning (DL) methods have become dominant in MRI-based knee injury detection.
Purpose of the Study:
- To systematically review deep learning applications for detecting knee injuries (anterior cruciate ligament, meniscus, cartilage) in MRI studies.
- To assess the performance and identify limitations of DL models in this domain.
Main Methods:
- Systematic literature review following PRISMA guidelines.
- Searches conducted across PubMed, Cochrane Library, EMBASE, and Google Scholar.
- Analysis of prediction accuracy and identification of common limitations.
Main Results:
- Deep learning models achieved prediction accuracies ranging from 72.5% to 100% for knee injury detection.
- DL shows potential to match human-level performance in MRI-based knee injury diagnosis.
- Identified limitations include data imbalance, model generalizability, verification bias, and ground-truth subjectivity.
Conclusions:
- Deep learning demonstrates significant potential for MRI-based knee injury diagnosis.
- Addressing limitations like explainability and lightweightness is key for clinical integration.
- Further research should focus on improving model robustness and generalizability for broader clinical application.
