A novel deep learning model for breast lesion classification using ultrasound Images: A multicenter data evaluation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer is a leading cause of cancer death in women, making early diagnosis critical for survival.
• Accurate classification of breast lesions is essential for effective screening and treatment.
• Breast biopsy, while the gold standard, is invasive and time-consuming.

Purpose of the Study:

• To develop a novel deep learning architecture for classifying ultrasound breast lesions.
• To enhance the InceptionV3 network by converting modules to residual inception and adjusting hyperparameters.
• To improve the accuracy and efficiency of breast cancer diagnosis.

Main Methods:

• Developed a deep learning model based on a modified InceptionV3 architecture.
• Utilized a combined dataset of five sources (three public, two private) for training and validation.
• Evaluated model performance using metrics including precision, recall, F1 score, accuracy, and AUC.

Main Results:

• The model achieved high performance on the test dataset with an accuracy of 0.81 and an AUC of 0.81.
• Key performance metrics included precision (0.83), recall (0.77), and F1 score (0.80).
• The Root Mean Squared Error was 0.18, indicating good model fit.

Conclusions:

• The enhanced InceptionV3 model demonstrates robust capability in classifying breast tumors.
• This AI-driven approach shows potential to decrease reliance on invasive breast biopsies.
• The study highlights the promise of deep learning in improving breast cancer diagnostic workflows.