Artificial Intelligence-Driven Quantification of Tumor-Stroma Ratio and Fibroblasts Enables Precise Classification of Stroma Quality and Quantity in Predicting Colorectal Cancer Recurrence
Jun Yong Kim, Hoein Jeong, Aaron Valero Puche, Sanghoon Song, Soo Ick Cho, Minsun Jung
Abstract
INTRODUCTION: The tumor microenvironment plays a crucial role in the progression and prognosis of colorectal cancer (CRC). Among its components, the tumor-stroma ratio (TSR) and cancer-associated fibroblasts (CAFs) have emerged as significant prognostic markers. However, conventional assessments of TSR and CAF density remain subjective and labor-intensive, limiting their clinical applicability. METHODS: We utilized an artificial intelligence (AI)-based whole slide image analysis platform, Lunit SCOPE IO, to objectively quantify TSR and CAF density in tissue samples from 207 treatment-naïve patients with stage II and III CRC. RESULTS: Our analysis demonstrated that both TSR (log-rank p < 0.0001) and CAF density (log-rank p = 0.017) were independently associated with disease-free survival (DFS). These AI-derived markers outperformed conventional prognostic factors. Furthermore, integrating TSR and CAF density with existing high-risk criteria enabled reclassification of additional patients as high risk, enhancing DFS prediction and reducing false-negative rates. CONCLUSION: AI-powered histopathological quantification of TSR and CAF density improves prognostic accuracy in CRC and offers a promising approach for refining risk stratification. These findings support the integration of AI-based pathology into clinical practice to enhance diagnostic precision and patient management.