Bridging the Gap Between Machine Learning and Medicine: A Critical Evaluation of the Dworak Regression Grade in Rectal Cancer

The growing popularity of artificial intelligence (AI) has increased its widespread adoption in medicine. However, the relationship between AI and medical experts’ opinions remains elusive. This study investigated the consistency between Random Forest’s prediction for rectal cancer regression grades and doctors’ opinion based on clinical data. We examined the impact of grading system subjectivity on the algorithm. Analyzing clinical parameters and medical notes from 85 rectal cancer patients, we identified patients with ambivalent grades, the “gray-zone patients,” and explored the algorithm’s difficulty in predicting their regression grade. We also introduced a regularization parameter to test if some patients could still correctly be predicted when some statistical information is suppressed. Our results demonstrated that the gray-zone patients were significantly more difficult to classify using the algorithm, suggesting that such patients should be reviewed twice to reduce errors. Additionally, we observed that the regularization parameter did not benefit gray-zone patients as much as others. Our findings emphasize the need for AI and clinical experts to work collaboratively since the algorithm cannot consider the subjectivity that medical experts can identify. Further research is necessary to incorporate subjectivity into AI algorithms to enhance their predictive capabilities and further bridge the gap between medicine and AI.