The knee is the most commonly imaged body part by MRI, and a new study has found that deep learning model can accelerate MRI knee exams while also improving their accuracy.
A team of researchers from the department of computer science at Stanford University set out to find if deep learning model could improve diagnostic accuracy for radiologists and orthopedic surgeons evaluating anterior cruciate ligament (ACL) tears, meniscal tears, and general abnormalities.
The model’s analysis was presented to seven general radiologists and two orthopedic surgeons who were asked to measure the “specificity, sensitivity, and accuracy” of the model’s findings. The physicians noted that the software improved detection of ACL tears by 4.8 percent, and had an area under the receiver curve (AUC) of .937 for abnormality identification and an AUC of .847 for meniscus tear detection.
The model can be used to streamline diagnostic workflows, enabling normal exams to be automatically labeled as “normal” in a preliminary reading. Although the model demonstrated successful results, the researchers note that their findings had some limitations. They only used MRI data from one facility, and a more diverse data set is needed in order to determine its level of accuracy on a more profound level.
“Our deep learning model can rapidly generate accurate clinical pathology classifications of knee MRI exams from both internal and external datasets,” write the study authors. “Moreover, our results support the assertion that deep learning models can improve the performance of clinical experts during medical imaging interpretation.”