Deep Learning Capabilities for the Categorization of Microcalcification.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Breast cancer is a leading global health concern, with microcalcifications being a key indicator on mammograms.
• Accurate classification of microcalcifications (benign, malignant, benign without callback) is crucial for timely diagnosis and treatment.
• Existing diagnostic methods can be labor-intensive and prone to inter-observer variability.

Purpose of the Study:

• To develop and evaluate an automated deep learning pipeline for the detection and classification of breast microcalcification categories.
• To compare the performance of Convolutional Neural Networks (CNNs) with different optimizers for microcalcification classification.
• To assess the efficacy of the proposed pipeline against established deep learning and classical machine learning techniques.

Main Methods:

• Utilized a deep learning approach with Convolutional Neural Networks (CNNs) for microcalcification detection and classification.
• Employed the pretrained InceptionResNetV2 model for feature extraction on 299 × 299 × 3 images.
• Tested the pipeline on the CBIS-DDSM mammogram dataset, evaluating performance using sensitivity, specificity, accuracy, and AUC.

Main Results:

• The proposed CNN pipeline successfully detected and classified three categories of microcalcifications.
• Performance was evaluated using various optimizers (ADAM, ADAGrad, ADADelta, RMSProp).
• The developed classification scheme demonstrated superior performance compared to previous deep learning and classical machine learning methods.

Conclusions:

• The automated deep learning pipeline offers a promising tool for accurate breast microcalcification classification.
• This approach has the potential to improve the efficiency and reliability of breast cancer diagnosis.
• Further research can explore larger datasets and diverse deep learning architectures for enhanced performance.