From cell counts to cellular interactions: Cu-Cyto and the co-localization index as a spatial framework for the tumor immune microenvironment of rectal cancer
7 affiliations
- 1Division of Gastrointestinal Surgery, Department of Surgery, Graduate School of Medicine, Kobe University, 7-5-2 Kusunoki-cho, Chuo-ku, Kobe, Hyogo 650-0017 Japan
- 2Division of Biophysics, Department of Health Sciences, Graduate School of Medicine, Kobe University, 7-10-2 Tomogaoka, Suma-ku, Kobe, Hyogo 654-0142 Japan
- 3Association of Medical Artificial Intelligence Curation (AMAIC), Sakae Members Office Building, Room 505, 4-16-8 Sakae, Naka-ku, Nagoya, Aichi 460-0008 Japan
- 4Division of Hepato-Biliary and Pancreatic Surgery, Department of Surgery, Graduate School of Medicine, Kobe University, 7-5-2 Kusunoki-cho, Chuo-ku, Kobe, Hyogo 650-0017 Japan
- 5Center for Medical Education and Clinical Training, Faculty of Medicine, Kindai University, 1-14-1 Miharadai, Minami-ku, Sakai, Osaka 590-0197 Japan
- 6Division of Radiation Oncology, Graduate School of Medicine, Kobe University, 7-5-2 Kusunoki-cho, Chuo-ku, Kobe, Hyogo 650-0017 Japan
- 7Department of Gastroenterological Surgery, Hyogo Cancer Center, 13-70 Kitaoji-cho, Akashi, Hyogo 673-8558 Japan
Abstract
The tumor immune microenvironment in colorectal cancer has emerged as a major determinant of patient outcome, with tumor-infiltrating lymphocytes, particularly CD8⁺ T cells, mediating antitumor immunity. Standardized density-based metrics such as the Immunoscore have demonstrated prognostic value but cannot capture the spatial interactions among cells that underlie immune–tumor biology. We introduce a deep learning framework that integrates two complementary axes of analysis: Cu-Cyto, a deep learning–based image cytometry platform that detects and classifies approximately twenty cell types from standard immunohistochemistry-stained whole-slide images, using a bit-pattern kernel-filtering algorithm to prevent multi-counting and an off-target labeling strategy for precise nuclear-center localization; and the Co-Localization Index, which converts the classification probabilities and coordinates produced by Cu-Cyto into a single quantitative measure of co-localization between two or three cell types. We illustrate this framework through CD103⁺CD8⁺ tissue-resident memory–like T cells in rectal cancer treated with neoadjuvant chemoradiotherapy: their stromal but not intratumoral density independently predicts relapse-free survival. This divergence, which compartment-aware density alone cannot explain, motivates a metric that captures within-compartment spatial relationships. The Co-Localization Index extends naturally to three-cell interactions, providing a quantitative readout of the tri-cellular biology among CD103⁺CD8⁺ T cells, tumor cells, and stromal components. Prospective validation in independent rectal-cancer cohorts, addressing watch-and-wait organ preservation and adjuvant chemotherapy decisions, will determine whether this spatial-cellular framework translates into routine clinical decision support with potential extension to other solid tumor contexts.
Related
- VISTA as a tumour-microenvironment checkpoint: spatial, metabolic and cellular control of antitumour immunity
- Lactylation-mediated remodeling of spatial immunometabolic niches in lung cancer: metabolic origins, cellular interactions, and molecular mechanisms
- Strategies of mechanical adaptations of circulating tumor cells and properties of their cellular microenvironment have predictive and prognostic biomarker value
- CNS localization drives a profoundly immunosuppressive microenvironment in B‑cell lymphoma
- Reprogramming the Melanoma Tumor Microenvironment: From Cellular Crosstalk to Cellular Immunotherapy
BibTeX
@article{yamashita2026from,
title = {{From cell counts to cellular interactions: Cu-Cyto and the co-localization index as a spatial framework for the tumor immune microenvironment of rectal cancer}},
author = {Kimihiro Yamashita and Toru Nagasaka and Tomoki Abe and Yukari Adachi and Ryota Ito and Takaaki Tachibana and Toru Takahashi and Hiroki Kagiyama and Masaki Imai and Takao Tsuneki and Masayuki Ando and Yuna Asano and Kumiko Miyashita and Yasunori Otowa and Naoki Urakawa and Hironobu Goto and Hiroshi Hasegawa and Shingo Kanaji and Mitsugu Fujita and Takeru Matsuda and Ryohei Sasaki and Takumi Fukumoto and Yoshihiro Kakeji},
journal = {International Journal of Clinical Oncology},
year = {2026},
month = sep,
doi = {10.1007/s10147-026-03182-0},
url = {https://doi.org/10.1007/s10147-026-03182-0},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - From cell counts to cellular interactions: Cu-Cyto and the co-localization index as a spatial framework for the tumor immune microenvironment of rectal cancer
AU - Kimihiro Yamashita
AU - Toru Nagasaka
AU - Tomoki Abe
AU - Yukari Adachi
AU - Ryota Ito
AU - Takaaki Tachibana
AU - Toru Takahashi
AU - Hiroki Kagiyama
AU - Masaki Imai
AU - Takao Tsuneki
AU - Masayuki Ando
AU - Yuna Asano
AU - Kumiko Miyashita
AU - Yasunori Otowa
AU - Naoki Urakawa
AU - Hironobu Goto
AU - Hiroshi Hasegawa
AU - Shingo Kanaji
AU - Mitsugu Fujita
AU - Takeru Matsuda
AU - Ryohei Sasaki
AU - Takumi Fukumoto
AU - Yoshihiro Kakeji
PY - 2026
DA - 2026/09/08
JO - International Journal of Clinical Oncology
DO - 10.1007/s10147-026-03182-0
UR - https://doi.org/10.1007/s10147-026-03182-0
AB - The tumor immune microenvironment in colorectal cancer has emerged as a major determinant of patient outcome, with tumor-infiltrating lymphocytes, particularly CD8⁺ T cells, mediating antitumor immunity. Standardized density-based metrics such as the Immunoscore have demonstrated prognostic value but cannot capture the spatial interactions among cells that underlie immune–tumor biology. We introduce a deep learning framework that integrates two complementary axes of analysis: Cu-Cyto, a deep learning–based image cytometry platform that detects and classifies approximately twenty cell types from standard immunohistochemistry-stained whole-slide images, using a bit-pattern kernel-filtering algorithm to prevent multi-counting and an off-target labeling strategy for precise nuclear-center localization; and the Co-Localization Index, which converts the classification probabilities and coordinates produced by Cu-Cyto into a single quantitative measure of co-localization between two or three cell types. We illustrate this framework through CD103⁺CD8⁺ tissue-resident memory–like T cells in rectal cancer treated with neoadjuvant chemoradiotherapy: their stromal but not intratumoral density independently predicts relapse-free survival. This divergence, which compartment-aware density alone cannot explain, motivates a metric that captures within-compartment spatial relationships. The Co-Localization Index extends naturally to three-cell interactions, providing a quantitative readout of the tri-cellular biology among CD103⁺CD8⁺ T cells, tumor cells, and stromal components. Prospective validation in independent rectal-cancer cohorts, addressing watch-and-wait organ preservation and adjuvant chemotherapy decisions, will determine whether this spatial-cellular framework translates into routine clinical decision support with potential extension to other solid tumor contexts.
SN - 1341-9625
SN - 1437-7772
ER -