Real-time breast lesion classification combining diffuse optical tomography frequency domain data and BI-RADS...

Area of Science:

• Medical Imaging
• Biomedical Optics
• Machine Learning in Healthcare

Background:

• Ultrasound (US)-guided diffuse optical tomography (DOT) shows promise for breast cancer diagnosis.
• Current DOT limitations include slow data processing and image reconstruction, hindering real-time applications.
• Accurate and rapid breast cancer diagnosis is crucial for effective patient management.

Purpose of the Study:

• To develop a real-time classification scheme for breast cancer diagnosis.
• To integrate diffuse optical tomography (DOT) measurements with Breast Imaging Reporting and Data System (BI-RADS) assessments.
• To improve the speed and accuracy of breast lesion classification.

Main Methods:

• A real-time classification scheme was developed combining US BI-RADS readings and DOT frequency domain measurements.
• A convolutional neural network (CNN) was trained to generate malignancy probability scores from DOT data.
• A support vector machine (SVM) classifier integrated CNN-derived scores with BI-RADS assessments for final diagnostic output.

Main Results:

• The combined US-BI-RADS and DOT approach achieved an area under the receiver operating characteristic curve (AUC) of 0.978.
• This method successfully distinguished between benign and malignant breast lesions using patient data.
• The classification was achieved without the need for image reconstruction, significantly improving processing speed.

Conclusions:

• The proposed real-time classification scheme effectively combines DOT and US-BI-RADS for accurate breast cancer diagnosis.
• This method overcomes the speed limitations of traditional DOT by eliminating the image reconstruction step.
• The high AUC demonstrates the potential of this integrated approach for rapid and reliable breast lesion assessment.