A Machine Learning Model Based on MRI Radiomics to Predict Response to Chemoradiation Among Patients with Rectal
Filippo Crimì1, Carlo D'Alessandro1, Chiara Zanon1
1Institute of Radiology, Department of Medicine-DIMED, University of Padova, 35128 Padova, Italy.
Life (Basel, Switzerland)
|January 8, 2025
Summary
Machine learning models using radiomics features from MRI can predict pathological complete response (pCR) to preoperative chemoradiotherapy (pCRT) in rectal cancer patients, aiding personalized treatment strategies.
Area of Science:
- Oncology
- Radiology
- Medical Imaging
- Machine Learning
Background:
- Rectum-sparing protocols are increasingly adopted for rectal cancer treatment.
- Predicting treatment response is crucial for optimizing patient management.
Purpose of the Study:
- To predict pathological complete response (pCR) to preoperative chemoradiotherapy (pCRT) in rectal cancer.
- To utilize pre-treatment MRI and radiomics with machine learning for response prediction.
Main Methods:
- 102 rectal cancer patients' MRI data were divided into training (n=72) and validation (n=30) cohorts.
- Machine learning models were trained using radiomic features to differentiate responders from non-responders.
- Histological results from total mesorectal excision determined pCR status.
Main Results:
- The best model achieved a ROC-AUC of 73% and 70% accuracy in the training cohort.
- In the validation cohort, the model demonstrated 81% sensitivity and 80% accuracy.
- The model showed a positive predictive value (PPV) of 80%.
Conclusions:
- Radiomics and machine learning show promise in predicting rectal cancer treatment response.
- These advanced methods support personalized rectal cancer management.
- Integration of imaging and computational techniques can enhance treatment strategies.
