A Geometrical Heuristic Image Processing Approach for the Automatic Detection & Quantification of Type-IV Crater Lesion Pathology in the Femoral Cartilage of the Human Knee

Abstract

The type-IV crater is a chondral lesion found in 5-10 percent of injured knees undergoing arthroscopy. The lesion is identified as a serious injury, being found predominantly in young adults, around 30 years of age, having the potential to quickly progress to osteoarthritis due to high load sporting activities typically pursued by this age group. In addition, pain and swelling often accompany larger lesions of this type, and it is recommended that treatment occurs as soon as possible while the lesion is well demarcated. The main limitation, in the quantification of type-IV crater volumes in 'real' human knees is that it is not possible to obtain the true volumes of the craters through a physical experimental measure and manual delineation is known to suffer from user error. We resolve this issue by performing extensive simulations of synthetic craters, of known radii and volume, within the femoral cartilage of 21 healthy knee joints to validate how a novel geometrical heuristic image processing approach can be used to detect type-IV crater lesions and quantify their volumes accurately in real-time from pre-segmented MR images which we compare to the standard manual delineation approach. We show that the mean %error in accuracy of our approach compared to the manual delineation approach for detecting and quantifying synthetic craters of 2-4 mm radii was significantly less (P<0.05) for our developed approach and the time was near instantaneous versus longer times required for manual delineation. Finally, we demonstrate how our developed approach could be used to detect type-IV crater pathology in a randomized sequence of 21 healthy and 4 real pathological knees (containing 5 type-IV craters in total). We show how our developed approach could identify real pathological from non-pathological knees with 100% accuracy (P<0.001). In addition, it was found that our developed approach could identify anatomical location of the real type-IV craters to be the same as a blinded operator's identification of the position of craters (Kappa=1). Furthermore, since our developed approach performed better than the standard manual delineation approach from the synthetic crater results, we applied it as the benchmark for the quantification of real type-IV crater volumes, and demonstrate how the manual delineation approach underestimated the real type IV crater volumes with a mean %error of 1.7% which was found to be consistent with the synthetic simulations. In conclusion, we demonstrate how a novel geometrical heuristic image processing approach can provide accurate real-time, automatic detection and quantification of type-IV crater lesions in pre-segmented MR images of the femoral cartilage for radii 2-5 mm. To the authors knowledge this is the first time type-IV crater lesions in the MR image of a human femoral cartilage have been detected and quantified automatically.