Deep Convolutional Neural Network-Based Lymph Node Metastasis Prediction for Colon Cancer Using Histopathological...

Area of Science:

• Oncology
• Pathology
• Computational Biology

Background:

• Accurate prediction of lymph node metastasis (LNM) is crucial for colon cancer treatment and follow-up.
• Current human evaluation of pathological slides has limitations in predicting LNM.
• There is a need for improved histopathological features for LNM prediction in colon cancer.

Purpose of the Study:

• To develop and validate accurate histopathological features for predicting LNM in colon cancer.
• To assess the predictive value of the peri-tumoral stroma (PTS) score for LNM.
• To establish a reproducible parameter for guiding colon cancer treatment decisions.

Main Methods:

• A deep convolutional neural network model was developed to analyze colon cancer tissue.
• The model was applied to whole-slide pathological images from The Cancer Genome Atlas (TCGA).
• The peri-tumoral stroma (PTS) score was assessed for its predictive value in LNM.

Main Results:

• Analysis of 164 TCGA patients with stages I-III colon cancer.
• Significantly higher PTS scores were observed in LNM-positive patients compared to LNM-negative patients (P < 0.001).
• PTS scores were significantly increased in cases with lymphatic invasion and other types of invasion in LNM-positive patients (P < 0.001).

Conclusions:

• The established PTS score is a simplified, reproducible parameter for predicting LNM in colon cancer.
• Computer-based analysis of the PTS score can aid in guiding treatment decisions for colon cancer.
• Further validation through large-scale prospective clinical trials is warranted.