Machine learning-based classification model to differentiate subtypes of invasive breast cancer using MRI

Area of Science:

• Oncology
• Radiology
• Biomedical Informatics

Background:

• Breast cancer, a leading cause of mortality in women, presents with distinct subtypes like Invasive Ductal Carcinoma (IDC) and Invasive Lobular Carcinoma (ILC).
• These subtypes exhibit variations in epidemiology, molecular characteristics, and clinical presentation, necessitating distinct diagnostic and treatment strategies.
• Accurate differentiation between IDC and ILC is crucial for effective patient management.

Purpose of the Study:

• To develop and validate a machine learning-based predictive model for distinguishing between IDC and ILC.
• To investigate the utility of contralateral breast morphological features derived from MRI in differentiating these breast cancer subtypes.
• To optimize the predictive model using Random Forest Classifier and hyperparameter tuning.

Main Methods:

• Acquisition of 143 magnetic resonance imaging (MRI) datasets from the DUKE Breast-Cancer collection.
• Extraction of contralateral breast morphological features using 3D Slicer software.
• Application of supervised machine learning, specifically a Random Forest Classifier, for IDC and ILC differentiation.

Main Results:

• The developed model achieved a classification accuracy of 79% and an Area Under the Curve (AUC) of 0.851.
• Key differentiating morphological features included contralateral breast volume, surface area, density, and the volume-to-surface area ratio.
• These findings highlight the potential of contralateral breast dimensions as significant indicators for differentiating IDC and ILC.

Conclusions:

• A robust machine learning model was successfully developed to differentiate IDC and ILC based on contralateral breast morphology.
• The study demonstrates the potential of non-invasive imaging features for improving breast cancer subtype classification.
• This approach offers a promising avenue for enhancing diagnostic accuracy and guiding personalized treatment strategies for breast cancer patients.