Computer aided detection (CAD) for meniscal and articular cartilages on magnetic resonance (MR) images

The injuries of the articular and meniscal cartilages of the knee are common in both young athletes and the aging population, requiring accurate diagnosis and, if appropriate, surgical intervention. With proper techniques and experience, the confidence in detection of meniscal tears and articular cartilage injuries can be very high. However, for many radiologists without musculoskeletal training, diagnosis of these cartilage injuries can be challenging. This dissertation develops CAD systems for automatic detection of meniscal tears and articular cartilage injuries of the knees to aid the radiologist in the diagnosis these injuries. For a CAD system to be used for automatic detection of meniscal cartilage, automated segmentation of sagittal T1-weighted MR image sequences of the knee and detection of tears is performed in two stages where the first stage consists of Region of Interest (ROI) selection, slice selection (automatic), binarization and enforcing shape constraints, while he second stage is a two-step process consisting of a scoring system that assigns scores to slices for potential tears using two newly introduced metrics and generation of the final recommendation regarding whether the meniscus is torn or normal. The evaluation is carried out on the 40 cases obtained from the University of Maryland, School of Medicine through an IRB (Institutional Review Board). The CAD application is able to detect tears in less than 15 seconds for all the cases and it gave sensitivity of 83% and specificity of 75%. This is comparable to average sensitivity of 87% and specificity of 81% by the two board certified musculoskeletal radiologists used in the evaluation of the CAD system. As for a CAD system designed for automatic detection of articular cartilage, a 2D Active Shape Model (ASM) is first used to model bone-cartilage interface on all the slices of the Double Echo Steady State (DESS) MR image sequences. It is then followed by measurements of the cartilage thickness from the surface of the bone. Finally, it outputs the identification of regions of abnormal thinness and focal/degenerative lesions. For articular cartilage injuries part, the evaluation is carried out on over 20 cases chosen from the Osteo-Arthritis Intitiative (OAI) database. Our CAD tool is able to fit the 2D ASM and correctly identify the slices containing the cartilage and obtain cartilage segmentation/thickness maps in little over 60 seconds for all of the cases. The cartilage maps generated by the CAD application were in good agreement with the interpretations of 2 board-certified musculoskeletal radiologists in over 85% of the cases. The feasibility of automatic detection of simple, complex meniscal tears and the detection of cartilage injuries of the knee in a near real time fashion using a CAD system could have immediate and beneficial clinical applications. The developed CAD system shows promise for increasing radiologist productivity and confidence, improving patient out comes, and applying more sophisticated CAD algorithms to orthopedic imaging tasks.