Machine learning-based classifiers to predict metastasis in colorectal cancer patients

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Colorectal cancer (CRC) presents a growing public health challenge in Iran, with increasing incidence over recent decades.
• Understanding and predicting metastasis is crucial for effective CRC patient management and treatment strategies.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting metastasis in colorectal cancer (CRC) patients.
• To identify key demographic and clinical factors influencing CRC metastasis prediction.

Main Methods:

• Utilized a dataset of 1,127 CRC patients, split into 80:20 training and testing sets.
• Applied various ML algorithms including Naive Bayes, Random Forest (RF), Support Vector Machine (SVM), Neural Network (NN), Decision Tree, and Logistic Regression.
• Evaluated model performance using 5-fold cross-validation, focusing on metrics like AUC, sensitivity, and specificity.

Main Results:

• Neural Network (NN) and Random Forest (RF) models demonstrated superior performance in predicting metastasis, achieving high AUC, sensitivity, and accuracy.
• SVM also showed strong predictive capabilities. Balanced datasets generally enhanced ML model performance.
• Tumor stage, lymph node count, and treatment type were identified as the most significant predictors of metastasis.

Conclusions:

• NN and RF are highly effective ML approaches for predicting metastasis in colorectal cancer (CRC) patients.
• Tumor stage and the number of involved lymph nodes are critical factors for metastasis prediction in CRC.