Introduction: Accurate assessment of lymph node metastasis risk post-endoscopic resection is crucial in T1 colorectal cancer, given the 10% incidence of metastases. Current predictive methods, such as histopathological factors and scoring systems, suffer from low inter-observer reliability and susceptibility to evaluator bias. We introduce an innovative artificial intelligence (AI)-assisted digital pathology system utilizing whole slide images (WSIs) to enhance the prediction of lymph node metastasis in T1 colorectal cancer.
Methods: We developed the AI model using 462 T1 colorectal cancer cases (65 with metastasis) surgically excised between 2001 and 2018. To assess the AI's performance, 100 consecutive T1 cancers resected from 2018 to 2021 served as validation cases. Pathological slides stained with hematoxylin and eosin were digitized into WSIs at 40x magnification using Nanozoomer technology. The WSIs were segmented into 224 x 224-pixel images and processed through weakly supervised learning via multiple instance learning (MIL). The resulting MIL scores were then integrated with patient sex and tumor location data, training a random forest classifier. We measured the model's efficacy by the area under the receiver operating characteristic (ROC) curve (AUC), comparing it to current clinical guidelines.
Results: In the validation cohort, lymph node metastasis was present in 15% of cases (15/100). The AI model achieved a higher AUC (0.73) than the guideline-based prediction (AUC=0.52, p<0.05), with both methods attaining 100% sensitivity. However, AI demonstrated superior specificity (35%) compared to the guideline (4.7%), potentially allowing for a 26% reduction in unnecessary surgeries without overlooking any cases of metastasis.
Conclusions: Our proposed AI system, leveraging WSI for T1 colorectal cancer, offers an objective and potentially more precise prediction of lymph node metastasis risk. This technology promises to curtail superfluous surgical interventions among T1 colorectal cancer patients at minimal metastatic risk.

Yushi Ogawa

Appearances