Finding what must not be missed in medical images. My research applies image processing and pattern recognition to support people when they make decisions.
I started in the 1990s with lesion detection in double-contrast X-ray images of the stomach and in X-ray CT images. Since then, the targets have moved to ultrasound videos and endoscopic video, and the methods from line-pattern concentration features and filter design to deep learning.
Beyond medicine, I work on a wide range of subjects that read human activity from images: sports form analysis, food photographs, time-series analysis of astronomical images, and character recognition of historical and modern official documents. Research with students is summarized on the Mekada & Doman Lab (MDLab) website (in Japanese).
Current Research
Liver tumor detection and tracking in abdominal ultrasound videos
Ultrasound can be repeated on the spot, but image quality varies widely and tumors are easily overlooked. Rather than detecting frame by frame, we aim at stable detection by following a tumor through the video, and study frameworks that learn detection and tracking jointly.
OCR error correction of historical documents with large language models
Modern historical materials such as the Taiwan Governor-General archives mix handwriting and print, and OCR output still contains errors that prevent search and analysis. We correct misreadings automatically, using context and the vocabulary specific to the documents.
Fundamental techniques for computer-aided diagnosis
Extraction of organs and lesions from CT images, surgical instrument segmentation in laparoscopic video, and registration of time-series images: the building blocks that make diagnostic support systems work.
Recent Publications
- Domain Adaptation from Visible-Light to FIR with Reliable Pseudo Labels
2023 18th International Conference on Machine Vision and Applications (MVA), 2023 - Artificial intelligence (AI) models for the ultrasonographic diagnosis of liver tumors and comparison of diagnostic accuracies between AI and human experts.
Journal of gastroenterology, 2022 - Detection and Tracking of Liver Tumors for Ultrasound Diagnostic Support Using Deep Learning
Journal of Image and Graphics, 2022 - Matching of sprint performance videos focusing on the motion periodicity and the speed linearity
IEEJ Transactions on Electrical and Electronic Engineering, 2021 - Lesion Image Generation Using Conditional GAN for Metastatic Liver Cancer Detection
Journal of Image and Graphics, 2021
Service
- Conference Chair, JAMIT 2026 (Japanese Society of Medical Imaging Technology)