Cartilage Segmentation from MRI Images Towards Prediction of Osteoarthritis

Mostly occurred arthritis in world-wide as well as in India is Osteoarthritis. Knee cartilage loss is the most preliminary and significant symptom for Osteoarthritis diagnosis. The automatic segmentation of knee cartilage from Magnetic Resonance Images (MRI) has a tremendous impact on research for diagnosis related to osteoarthritis. Even though deep learning approaches achieve success in this field, but effectively and accurately segmenting knee cartilage is challenging due to the complex and overlapping structure of the soft tissues in the MRI images. Thus it is necessary to study the performance of the existing conventional methods implementation in knee cartilage segmentation from MRI. This can infer the challenges of the existing segmentation methods which can be solved by the recent deep learning based segmentation concept towards building a new generalized segmentation model. Depending upon this phenomenon, we have investigated the state-of-the-art conventional and deep learning based segmentation methods for knee cartilage segmentation using MRI images. After that, we have also introduced an encoder-decoder framework inspired from U-Net segmentation architecture for solving the knee cartilage segmentation challenges. Experimental results reveal that the proposed encoder-decoder framework inspired from U-Net architecture is well-performed as compared to the state-of-the-art segmentation methods for cartilage segmentation with dice similarity index value of 0.905 using publicly available OAI dataset.