The KMeansGraphMIL Model: A Weakly Supervised Multiple Instance Learning Model for Predicting Colorectal Cancer Tumor...

Area of Science:

• Oncology
• Computational Biology
• Biomedical Imaging

Background:

• Colorectal cancer (CRC) is a leading cause of cancer-related deaths globally.
• Immunotherapy checkpoint blockade treatments show promise for CRC but require specific biomarker measurements.
• Tumor mutational burden (TMB) is a key biomarker, but traditional measurement via next-generation sequencing is costly and time-consuming.

Purpose of the Study:

• To develop an economical and rapid method for predicting TMB in colorectal cancer patients.
• To introduce the KMeansGraphMIL model for weakly supervised multiple-instance learning in TMB prediction.

Main Methods:

• Proposed the KMeansGraphMIL model utilizing weakly supervised multiple-instance learning.
• Incorporated both image patch feature vector similarity and spatial relationships.
• Evaluated model performance against previous weakly supervised multiple-instance learning approaches.

Main Results:

• Achieved an area under the receiver operating characteristic curve of 0.8334.
• Significantly improved recall to 0.7556.
• Demonstrated superior performance compared to existing weakly supervised multiple-instance learning models.

Conclusions:

• The KMeansGraphMIL framework offers an economical and rapid approach for predicting CRC TMB.
• This method has the potential to expedite physician treatment planning.
• The study presents a significant advancement in cost-effective and timely TMB assessment for colorectal cancer management.