Anomaly-Aware 3D Segmentation of Knee Magnetic Resonance Images

Boyeong Woo, Craig Engstrom, Jurgen Fripp, Stuart Crozier, Shekhar S. Chandra

Show abstract - Show schedule - Proceedings - PDF - Reviews

Automated segmentation of 3D medical images involves numerous challenges including factors such a lack of large image datasets for training in machine learning approaches, differences in image characteristics within and across imaging technologies and anatomical variations across individuals. Further challenges arise in clinical investigations when images contain incidental or coexisting anomalies alongside the primary pathoanatomy under examination. Therefore, successful automated segmentation of objects in the presence of such anomalies represents an important task in medical image analysis. In this work, we show how popular U-Net-based neural networks can be used for detecting anomalies in the cancellous bone of the knee from 3D magnetic resonance (MR) images in patients with varying grades of osteoarthritis (OA). We also show that the extracted information can be utilised for downstream tasks such as parallel segmentation of anatomical structures along with associated anomalies such as bone marrow lesions (BMLs). For anomaly detection, a U-Net-based model was adopted to inpaint the region of interest in images so that the anomalous regions can be replaced with close to normal appearances. The difference between the original image and the inpainted image was then used to highlight the anomalies. The extracted information was then used to improve the segmentation of bones and cartilages; in particular, the anomaly-aware segmentation mechanism provided a significant reduction in surface distance error in the segmentation of knee MR images containing severe anomalies within the distal femur. Furthermore, all of these U-Net-based models were fully volumetric convolutional neural networks, allowing for efficient 3D image processing.
Hide abstract

Wednesday 6th July
Poster Session 1.1 - onsite 15:20 - 16:20, virtual 11:00 - 12:00 (UTC+2)
Hide schedule