Feasibility of the fat-suppression image-subtraction method using deep learning for abnormality detection on knee MRI
Shusuke Kasuya1, Tsutomu Inaoka1, Akihiko Wada2
1Department of Radiology, Toho University Sakura Medical Center, Sakura, Japan.
Polish Journal of Radiology
|February 16, 2024
Summary
A deep learning model can generate quality fat-suppression images for knee MRI, enabling abnormality detection via image subtraction. This AI approach shows high accuracy in identifying various knee conditions.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Knee magnetic resonance imaging (MRI) is crucial for diagnosing joint disorders.
- Fat-suppression techniques enhance lesion visibility but can be time-consuming.
- Developing automated methods for generating and analyzing MRI sequences is an active research area.
Purpose of the Study:
- To assess the feasibility of a deep learning (DL) model in generating fat-suppression (FS) knee MRI images.
- To evaluate the DL model's capability in detecting abnormalities using the FS image-subtraction method.
Main Methods:
- A DL model using 2D convolutional neural networks was developed.
- The model generated FS images and performed subtraction between normal and abnormal scans.
- Image quality and detection performance metrics (accuracy, precision, recall, F-measure, sensitivity, AUROC) were calculated.
Main Results:
- Generated FS images were of adequate quality.
- 89.1% of subtraction images were deemed adequate.
- High accuracies (89.5-95.1%) were achieved for detecting various abnormalities, including ACL, bone marrow, cartilage, and meniscus.
Conclusions:
- The DL model successfully generated sufficient-quality FS images for knee MRI.
- The FS image-subtraction method, powered by DL, effectively detects abnormalities.
