Uncovering Hidden Prognostic Patterns in Colorectal Cancer Histology Using Unsupervised Learning: A Computational...

Area of Science:

• Computational pathology
• Oncology
• Artificial intelligence in medicine

Background:

• Colorectal cancer (CRC) is a major global health concern with limited diagnostic features from current histopathology.
• There's a need for advanced methods to identify subtle prognostic indicators in CRC tissues.

Purpose of the Study:

• To discover prognostically significant histomorphological patterns in CRC using unsupervised deep learning.
• To develop and validate a prognostic risk model integrating novel morphological features with clinical data.

Main Methods:

• Developed a framework combining convolutional neural networks and deep clustering on 23,341 CRC tissue image patches.
• Identified 30 distinct histomorphological clusters and performed survival analyses.
• Integrated significant clusters with clinical factors (T stage, N stage, differentiation) to build a risk model.

Main Results:

• Identified three histomorphological clusters (Cluster13, Cluster19, Cluster24) significantly associated with patient prognosis.
• The prognostic risk model stratified patients into high-risk and low-risk groups with significant survival differences in training and validation sets.
• Incorporating morphological clusters improved predictive performance modestly but significantly compared to clinical factors alone.

Conclusions:

• Computational pathology can reveal novel, visually subtle morphological features with independent prognostic value in CRC.
• These findings offer potential to refine CRC patient stratification and enhance clinical decision-making.
• The study highlights the complementary role of AI-driven histomorphology in oncology diagnostics.