(F-217) Color Space Based Segmentation Method For Spatial Analysis of Cytotoxic T Lymphocytes In Colorectal Cancer Patients

The American Cancer Society estimates 153,020 new cases and 52,550 deaths from Colorectal Cancer (CRC) in 2023. Treatment strategies for CRC patients place heavy emphasis on CRC staging. For Stage I CRC patients, surgical resection is the primary curative treatment. Patients in Stage 3 and 4 usually undergo surgical resection followed by adjuvant therapy, while therapy is only recommended for high-risk Stage II patients. Determination of high-risk, however, remains subjective. Furthermore, patients across all stages undergo relapse or CRC-related death, even though it is relatively rare in Stage I and II CRC patients. We hypothesize that characterizing CD8+ cytotoxic lymphocyte infiltration has objective prognostic value for identifying colorectal cancer patients truly at risk of relapse across all stages. As a first step toward validating our hypothesis we have developed a new segmentation algorithm capable of accurately locating CD8+ T cells in whole slide immunohistochemistry tissue images. Our algorithm does not require any training data.

Immunohistochemistry whole slide images of resected colorectal cancer sections from over hundred patients were stained and imaged. Colorectal cancer samples were acquired from over a hundred colorectal cancer patients at varying cancer stages.

            Segmentation method

After removing small artifacts from the images using median filtering, we performed an RGB to LAB space transformation. This transformation allowed us to capture the CD8+ brown stain. We performed contrast stretching and difference of Gaussian filtering to clearly define the edges of the stained cells. Finally multi-Otsu thresholding is performed to segment the cells.

Results

            Below we show a range of images where our algorithm is accurately able to perform CD8+ T cell segmentation. Furthermore, we performed Dirichlet tessellation based spatial processing on a microsatellite instable (MSI) high and an MSI low patient. We expect that the CD8+ T cells will be more densely clustered in MSI high patient with respect to the MSI low patient. This is exactly what our analysis revealed, giving first evidence of the applicability of our method in future analysis.

[Figure 1]

[Figure 2]

[Figure 3]

Discussion

            The goal of this research was to develop a segmentation algorithm to eventually characterize CD8+ T cell infiltration. We have developed such an algorithm, we aim to utilize this algorithm in the next steps of our work to accurately and objectively predict risk of cancer relapse in individual colorectal cancer patients across their MSI and cancer stage axes.