Enhancement and automated segmentation of ultrasound knee cartilage for early diagnosis of knee osteoarthritis

Abstract

Ultrasound(US) has emerged as a valid imaging modality for quantitative assessment of femoral cartilage thickness for early diagnosis of knee osteoarthritis (OA). In this work, we are presenting a framework for automated segmentation of knee cartilage from enhanced US images. The proposed framework involves enhancement of US bone surfaces by calculating local phase image features, dynamic programming for bone segmentation and the use of segmented bone surfaces as initial seeds to random walker (RW) algorithm. Qualitative and quantitative validation was performed on 100 scans obtained from eight healthy volunteers. Validation against expert manual segmentation achieved a mean dice similarity coefficient (DSC) of 0.8758.