A Comprehensive Review of Knee Abnormalities Detection Current Procedures, Outcomes and Prospects

Knee joints are made up of cartilage, bone, fluid and ligaments in which muscles and tendons are there for helping knee joints movement. Knee pains caused due to injury, repeated stress on knee or aging. Knee abnormalities or injury can be diagnosed using different methodologies like X-RAY, CT scan, Arthroscopy or MRI. In our survey, we thoroughly analyzed several recent research to investigate the landscape of predictive modeling, a comprehensive analysis of current literature, evaluating trends, and methodologies. We identified that deep-learning models showcased average accuracy for predicting that knee is normal or abnormal is ranged from 63.4% to 98%. In this review paper, we also highlight several limitations like challenges for generalization of models in different centers, biasness in verification, lack of multi-classification studies, unavailability of data and subjectivity of ground-truth. We also proposed An Explainable Attention Based Deep Learning Knee Anomalies Prediction Model with Ensemble Learning Algorithms framework. Here, multi-modal deep learning model architecture is designed to handle and learn from multiple types of data inputs. We achieved 84.54% accuracy with Kaggle knee dataset.