The histological evaluation of colorectal cancer (CRC – Colorectal Cancer) is a cornerstone of patient treatment planning. Traditionally, pathologists manually scan tissue sections to visually identify the boundaries of features such as adenocarcinoma invasion, stromal involvement, and different levels of dysplasia. However, this process is labor-intensive and subject to inter-observer variability — different experts may disagree on the exact visual boundaries of the same regions.

Our AI-driven approach mitigates human fatigue and subjectivity by providing a standardized, reproducible segmentation map across the entire histological slide. By automating the visual delineation of stroma, normal glands, and malignant or dysplastic regions, the system acts as an objective reference guide. It highlights critical regions of interest immediately, allowing clinicians to focus their time on high-level diagnostic synthesis rather than exhaustive manual scanning.

The service applies a deep learning pipeline for the automated morphological segmentation of colorectal cancer within Whole Slide Images (WSI). Segmentation is performed at the pixel level, visually mapping the tumor microenvironment and glandular structures. The result is a heatmap overlay displayed directly on the slide, accessible through a dedicated collaborative digital pathology environment.

What the service analyzes

The AI system automatically processes the histological image and provides:

• Automated tissue detection — automatically identifies and masks the relevant tissue area on the slide, ignoring background artifacts to optimize the inference process
• Multi-class segmentation — classifies tissue pixels into one of five distinct histological categories: stroma, adk_invasion (adenocarcinoma), high_grade, low_grade, and normal_gland
• Collaborative environment — results are available on a dedicated digital pathology platform (Cytomine), enabling real-time remote review, slide navigation, and multi-user discussion

Who this is for

This service is intended for:

• Pathologists and gastroenterologists who want to efficiently identify key histological regions and reduce the manual workload of scanning slides
• Clinical researchers needing a standardized visual segmentation of CRC tissue classes for collaborative study and systematic annotation

Input data

The service accepts digital pathology images in Whole Slide Image (WSI) — .svs format.

WSI files are high-resolution histological images obtained by digitally scanning biopsy or resection slides.

Results

The AI model processes the histological slide and generates an interactive heatmap — a color-coded overlay mapped directly onto the WSI, visually representing the five tissue classes detected by the model. Each region of the slide is assigned a color corresponding to its identified histological category, providing an immediate overview of the spatial distribution of tumoral and glandular structures.

Results are accessible through a web-based digital pathology visualizer, which supports seamless slide navigation (zoom/pan), toggling the heatmap on and off, manual annotation, note-taking, and collaboration with other users on the platform.

How to access the service

The service runs on a dedicated digital pathology platform (Cytomine), accessible through your institutional account using the same credentials as the RNCC platform (Keycloak).

Usage workflow:

• Access the platform using the button below
• Log in using the admin user with the password password
• Upload your WSI file in .svs format
• The system automatically runs tissue detection, preprocessing, and model inference
• Once processing is complete, navigate the slide, explore the heatmap, make annotations, and collaborate with other platform users