DEEP LEARNING BASED HKA ANGLE ASSESSMENT FOR DETECTION OF CHANGES IN KNEE JOINT ALIGNMENT

Abstract

Changes in knee alignment takes place with the occurrence of Knee Osteoarthritis. Malalignment of knee is evident when Hip knee ankle angle of an individual changes from its neutral position. The aim of the research is to study the lower-extremity using deep learning for detection of changes in joint alignment. In this study, a dataset of 340 images of scanograms was created. A deep learning model based on Renet-18 with transfer learning is implemented for analysis of left leg and right leg separately. Hip knee ankle angle is measured using landmarks on centre of Femur, centre of knee and centre of ankle. Depending on the HKA angle, knee alignment is decided as Varus, Valgus or Neutral. Results obtained by our deep learning model differed from radiologist’s predictions by ±2° for right leg and ±3° for left leg. The model could achieve Intra Class Correlation Coefficient (ICC) of 0.92 and Pearson Correlation Coefficient (r) of 0.94 in HKA analysis. Along with HKA angle analysis, model is trained to evaluate leg length and discrepancy between left and right leg. This deep learning technique enables precise predictions of HKA and leg length enabling the healthcare providers for early and fast decision making.