A random initialization deep neural network for discriminating malignant breast cancer lesions

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Breast cancer is a leading cause of death in women globally.
• Accurate early detection is crucial for effective treatment and improved survival rates.
• High rates of false negatives in diagnosis can lead to delayed treatment and adverse health outcomes.

Purpose of the Study:

• To design and validate a specialized convolutional neural network (CNN) architecture for breast lesion classification.
• To explore parameter combinations and architecture styles for optimal model selection.
• To develop a criterion emphasizing the reduction of false negatives while maintaining high accuracy in breast cancer diagnosis.

Main Methods:

• Development of an ad-hoc CNN architecture tailored for breast lesion classification.
• Heuristic exploration of various parameter combinations and architectural styles.
• Validation of the proposed CNN model on independent datasets to assess performance.

Main Results:

• The developed CNN architecture achieved good classification performance on validation and test sets.
• The model demonstrated practical utility in classifying and staging breast cancer.
• The study highlights the potential of custom CNNs in improving diagnostic accuracy and reducing false negatives.

Conclusions:

• An ad-hoc CNN architecture can significantly aid in breast cancer classification and staging.
• The proposed model selection criterion helps prioritize the reduction of false negatives.
• Deep learning offers a promising avenue for enhancing computer-assisted diagnosis in oncology.