Machine Learning techniques in breast cancer prognosis prediction: A primary evaluation.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning Applications

Background:

• Over 750,000 women in Italy are breast cancer survivors, highlighting the need for precise prognostic tools.
• Predicting individual disease trajectories and tailoring treatment remains complex despite existing prognostic factors.
• Multidisciplinary approaches and tools like Multigene Signature Panels and the Nottingham Prognostic Index are utilized but have limitations.

Purpose of the Study:

• To conduct a primary evaluation of machine learning (ML) applications for predicting breast cancer prognosis.
• To assess the efficacy of ML models in forecasting cancer recurrence and disease-specific mortality.
• To explore the potential of ML as an adjunct resource for clinical decision-making in breast cancer care.

Main Methods:

• Analysis of data from 610 patients who underwent surgery for breast cancer.
• Development of two types of ML models: Artificial Neural Network and Support Vector Machine.
• Prediction of three key outcomes: loco-regional recurrence, systemic recurrence, and death from disease within 32 months.

Main Results:

• ML models demonstrated high accuracy, ranging from 95.29% to 96.86%.
• Specificity was notably high (0.97-0.99), while sensitivity varied (0.35-0.64).
• Area Under the Curve (AUC) values ranged from 0.804 to 0.916, indicating good predictive performance.

Conclusions:

• The developed ML models show encouraging results in predicting breast cancer prognosis, with high specificity and accuracy.
• These models could serve as a valuable additional resource for clinicians in daily practice for patient prognosis.
• Further improvements in sensitivity are needed, potentially by incorporating larger patient cohorts and longer follow-up periods.