Novel Breast Imaging and Machine Learning: Predicting Breast Lesion Malignancy at Cone-Beam CT Using Machine Learning...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology Diagnostics

Background:

• Breast cancer diagnosis relies on imaging, with cone-beam CT (CBCT) offering a specialized modality.
• Evaluating the efficacy of advanced computational methods for improving diagnostic accuracy in breast imaging is crucial.

Purpose of the Study:

• To assess the diagnostic capabilities of various machine learning (ML) techniques for predicting malignancy in breast lesions detected via CBCT.
• To compare the performance of these ML models against experienced human readers.

Main Methods:

• Five ML algorithms (random forests, BPN, extreme learning machines, SVM, K-nearest neighbors) were trained and validated on a clinical breast CBCT dataset.
• Diagnostic performance was evaluated using area under the curve (AUC), sensitivity, and specificity.
• Two experienced radiologists independently analyzed the same dataset for comparison.

Main Results:

• Back propagation neural networks (BPNs) achieved the highest diagnostic performance among ML techniques, with an AUC of 0.91, sensitivity of 0.85, and specificity of 0.82.
• BPN performance significantly outperformed both human readers (p < 0.01 for reader 1, p < 0.001 for reader 2) in AUC and specificity.
• Human reader 1 achieved an AUC of 0.84, while reader 2 had an AUC of 0.72.

Conclusions:

• Machine learning techniques offer robust and high diagnostic performance for identifying malignancy in breast lesions on CBCT scans.
• BPNs emerged as the leading ML method, demonstrating superior diagnostic accuracy over human interpretation in this study.
• These findings suggest ML's potential to enhance the accuracy and reliability of breast cancer diagnosis using CBCT.