← All research themes Active learning Active learning and information acquisition We study which images or image pairs to annotate, who should provide the annotations, and how detailed they should be under a limited budget. For domain adaptation in pathology, we select informative images from a new site for annotation.
Spatial transcriptomics Cell image analysis and generation Vascular and multimodal medical imaging Pathology and endoscopic image analysis Label-efficient learning Active learning and information acquisition Medical AI under domain shift Foundations and other research Selected Papers
Selected papers Fig. 1 · Combining human and VLM weak annotations · Source: paper PDF ↗ ICIP 2026 · VLM as a weak annotator
Leveraging Vision-Language Models as Weak Annotators in Active Learning Human fine-grained labels and VLM-generated coarse labels are allocated under an annotation budget.
Paper and abstract ↗ Fig. 1 · Optimizing annotation granularity · Source: paper PDF ↗ Selected CVPR 2025 · Supervision-level selection
Instance-wise Supervision-level Optimization in Active Learning The method chooses the supervision level for each image to improve learning within a labeling budget.
Fig. 2 · Selecting pairs for relative annotation · Source: paper PDF ↗ Selected Medical Image Analysis 2024 · Endoscopy · Learning to rank
Deep Bayesian Active Learning-to-Rank with Relative Annotation for Estimation of Ulcerative Colitis Severity The method selects informative image pairs using model uncertainty and relative severity annotations.
Paper and abstract ↗ Cross-topic research
Active domain adaptation This work also addresses domain shift in pathology images.
Fig. 2 · Selecting samples by cluster entropy · Source: paper PDF ↗ ISBI 2023 · Pathology · Active domain adaptation
Cluster Entropy: Active Domain Adaptation in Pathological Image Segmentation The method selects informative slides for additional annotation when pathology images shift between sites. Also listed under domain adaptation.
Paper and abstract ↗ Publications
Related publications Phuong Ngoc Nguyen, Kaito Shiku, Ryoma Bise, Seiichi Uchida, and Shinnosuke Matsuo · IEEE International Conference on Image Processing (ICIP), 2026. (Spotlight Oral; top 8% of accepted papers)
Shinnosuke Matsuo, Riku Togashi, Ryoma Bise, Seiichi Uchida, and Masahiro Nomura · IEEE CVPR, 2025. (Top Conference in Computer Vision, acceptance rate:22.1%)
Takeaki Kadota, Hideaki Hayashi, Ryoma Bise, Kiyohito Tanaka, and Seiichi Uchida · Medical Image Analysis, 2024. (accepted, IF:10.7)
Xiaoqing Liu, Kengo Araki, Shota Harada, Akihiko Yoshizawa, Kazuhiro Terada, Mariyo Kurata, Naoki Nakajima, Hiroyuki Abe, Tetsuo Ushiku, and Ryoma Bise · IEEE International Symposium on Biomedical Imaging (ISBI), 2023. (Oral)
View all publications →
