Automated deep learning accurate in detecting knee joint damage
An automated deep learning-based system can accurately evaluate knee joint cartilage to detect wear and injury, according to a recent Radiology study.
The research—led by Fang Liu, with the University of Wisconsin School of Medicine and Public Health—may also lead to reduced reader variability and improved patient care.
Fang and colleagues utilized segmentation and classification convolutional neural networks (CNNs) to train the automated deep learning cartilage lesion detection system.
To test the deep learning model, the team used retrospective data sets from 175 patient who underwent fat-suppressed T2-weighted fast spin-echo MRI. The reference standard for training the CNN classification was based on prior musculoskeletal radiology interpretation of the articular surfaces of the femur and tibia.
Two separate evaluations were performed using the receiver operating curve (ROC) to measure diagnostic performance and in determining intraobserver agreement.
Results showed a ROC of 0.917 and 0.914 for evaluation one and two, respectively, “indicating high overall diagnostic accuracy for detecting cartilage lesions,” the group wrote.
After evaluation one, the deep learning system achieved 84.1 percent sensitivity and 85.2 percent specificity. The second evaluation resulted in 80.5 percent sensitivity and 87.9 percent specificity. Both sets were comparable to trained radiology experts, according to authors.
“The high sensitivity of the cartilage lesion detection system is particularly favorable, because the main limitation of MRI for evaluating articular cartilage (even by experienced musculoskeletal radiologists) is its relatively low sensitivity for identifying superficial cartilage lesions,” they wrote.
The lesion detection system also offered greater intraobserver agreement compared to the interobserver agreement demonstrated by clinical radiologists.
Liu and colleagues expressed cautious optimism, suggesting although their results are a step in the right direction, the study merely addressed the feasibility of using the automated system.
“Our study demonstrated the feasibility of using a fully automated deep learning–based cartilage lesion detection system for evaluating the articular cartilage of the knee joint,” the authors wrote. “While our initial results are promising, future work is needed for further technical development and validation of the cartilage lesion detection system before it can be fully implemented in clinical practice.”
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.