Development of Temporal Subtraction Technique for Phalanges CR Image using Geometric-matching CNN

• Published in: International Conference on Control, Automation and Systems (Online) pp. 558 - 561
• Main Authors: Ono, Hikaru, Kamiya, Tohru, Aoki, Takatoshi
• Format: Conference Proceeding
• Language: English; Japanese
• Published: ICROS 27.11.2022
• Subjects: Arthritis; Computer-Aided Diagnosis; Control systems; Convolutional neural networks; Data visualization; Image Registration; Image segmentation; Rheumatoid Arthritis; Subtraction techniques; Temporal Subtraction; X-rays
• ISSN: 2642-3901
• DOI: 10.23919/ICCAS55662.2022.10003828

We are developing a computer-aided diagnosis system for rheumatoid arthritis. X-rays images are widely used to diagnose the rheumatoid arthritis. However, it is difficult for physicians to read minute changes from the images. Therefore, we propose a method to visualize lesions in the phalangeal region by comparing past and current images of the same subject using a temporal subtraction technique. The proposed method consists of three steps: segmentation of phalanges, registration, and generation of subtraction images. First, the phalangeal region is extracted from the hand CR image using DeepLabv3+. Next, the past and current phalangeal region images are aligned by geometric-matching based on a CNN (convolutional neural networks) with instance-specific optimization. Finally, we apply the temporal subtraction technique to those images. We confirmed the effectiveness of the proposed registration method in an experiment using synthetic data. Also, the proposed method was applied to a pair of past and current image sets on same subject to generate a subtraction image. As a result, we confirmed that the proposed method can visualize changes between past and current images.