KEY TAKEAWAYS
- The study aimed to develop dosiomics-based ML models for high-risk individuals to predict AST, facilitating tailored treatment plans for high-risk individuals.
- The study found dosiomics-based ML effective in AST prediction, with improved accuracy when combined with DVH and PTR.
Acute skin toxicity (AST) induced by radiation is a prevalent side effect of breast cancer radiation therapy.
Pegah Saadatmand and the team aimed to develop machine learning (ML) models based on dosiomics to predict AST, facilitating tailored treatment plans for high-risk individuals.
The study utilized dosiomics features extracted via Pyradiomics (v3.0.1) alongside treatment plan-derived dose volume histograms (DVHs) and patient-specific treatment-related (PTR) data from breast cancer patients. Clinical scoring utilized the Common Terminology Criteria for Adverse Events (CTCAE) V4.0 for skin-specific symptoms.
About 52 breast cancer patients were grouped into AST 2+ (CTCAE ≥ 2) and AST 2- (CTCAE < 2) toxicity grades for modeling. They were randomly split into training (70%) and testing (30%) cohorts. Various prediction models were evaluated via multivariate analysis, combining dosiomics, DVH, and PTR features.
The 7 unique combinations, alongside 7 classification algorithms, were considered. Model performance was assessed on the test group using the area under the receiver operating characteristic curve (AUC) and f1-score. Additionally, each model’s accuracy, precision, and recall were examined. Statistical analysis involved feature differences between AST 2- and AST 2+ groups, including cutoff value calculations.
The results indicated that 44% of patients developed AST 2+ following Tomotherapy. The dosiomics (DOS) model, utilizing dosiomics features, notably improved the AUC (up to 0.78) when preserving spatial information in dose distribution compared to DVH features (up to 0.71). Additionally, a baseline ML model using only PTR features highlighted the significance of dosiomics in early AST prediction. Utilizing Extra Tree (ET) classifiers, the DOS + DVH + PTR model demonstrated statistically significant improved performance in AUC (0.83; 95% CI 0.71-0.90), accuracy (0.70), precision (0.74), and sensitivity (0.72) compared to other models.
This study provided a practical tool for healthcare professionals by demonstrating the effectiveness of dosiomics-based ML in predicting AST and the significant enhancement when combined with DVH and PTR. These findings open up avenues for timely interventions to prevent radiation-induced AST.
Funding was provided by the Iran University of Medical Sciences.
Source: https://pubmed.ncbi.nlm.nih.gov/38735974/
Saadatmand, P., Mahdavi, S.R., Nikoofar, A. et al. (2024). “A dosiomics model for prediction of radiation-induced acute skin toxicity in breast cancer patients: machine learning-based study for a closed bore linac.” Eur J Med Res. 2024 May 12;29(1):282. doi: 10.1186/s40001-024-01855-y. PMID: 38735974.
