Isao Matsui, Ayumi Matsumoto, Atsuhiro Imai, Hiroki Okushima, Hirohiko Niioka, Masatoshi Abe, Natsune Tamai, Hajime Nagasu, Eiichiro Kanda, Eiichiro Uchino, Tadashi Sofue, Toshiyuki Imasawa, Yuichiro Yano, Hiroshi Kinashi, Ken‐ichi Miyoshi, Tamaki Harada, Yasuyuki Nagasawa, Keiji Fujimoto, Yuka Kurokawa, Sawako Kato, Ryohei Kaseda, Masahiro Koizumi, Yasuo Kusunoki, Masaki Ohya, Yoshimasa Kawazoe, Hiroyuki Abé, Yuta Matsukuma, Takaaki Kosugi, Yoshiyasu Ueda, Naohiko Fujii, Masanobu Takeji, Akira Suzuki, Katsuyuki Nagatoya, Kazumasa Oka, Yutaka Ando, Masaaki Izumi, Toshiyuki Komiya, Tatsuo Tsukamoto, Imari Mimura, Takahiro Kuragano, Toshiaki Nakano, Kazuhiko Tsuruya, Yasuhiko Ito, Tetsuo Minamino, Osamu Yamaguchi, Suguru Yamamoto, Hirotaka Komaba, Kengo Furuichi, Kei Fukami, Shin‐ichi Araki, Takao Masaki, Naotake Tsuboi, Hitoshi Yokoyama, Akira Shimizu, Tetsuo Ushiku, Shoichi Maruyama, Motoko Yanagita, Masaomi Nangaku, Ryohei Yamamoto, Kazunori Inoue, Yoshitaka Isaka
Artificial intelligence for rare pathological lesion detection faces dual challenges: expert annotation scarcity and domain shifts across institutions. Using multi-institutional kidney biopsies from 22 hospitals with 3 scanner types (NDPI, VSI, SVS), we demonstrate that model performance decreases dramatically across domains, with up to 70.3% reduction in detection precision for rare lesions such as crescents and segmental sclerosis (comprising only 2-3% of annotations). We present an approach integrating semi-supervised learning with residual CycleGAN-based domain adaptation, reducing mean Fréchet inception distance between institutions from 55.9 to 20.2 while preserving diagnostic morphology. We identified context-dependent optimal strategies: semi-supervised learning with 50% confidence threshold excelled in same-hospital scenarios (15.2-17.7% improvement for rare lesions), while our combined GAN-Semi-Supervised approach demonstrated superior performance in cross-scanner scenarios between NDPI and VSI formats (up to 63.4% improvement for crescents). This methodology enables robust performance across diverse healthcare settings with minimal expert annotation.