Spatially localized sparse approximations of deep features for breast mass characterization

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Machine Learning

Background:

• Breast cancer is a leading cause of mortality globally.
• Accurate diagnosis of breast masses in mammograms is crucial for effective treatment.
• Deep learning shows promise in computer-aided diagnosis but requires further study.

Purpose of the Study:

• To investigate the use of deep-feature-generated dictionaries for sparse approximation-based classification of breast masses.
• To develop and evaluate a novel deep feature-based sparse approximation classification technique.
• To enhance the accuracy of classifying breast masses as benign or malignant in mammograms.

Main Methods:

• Proposed a deep feature-based sparse approximation classification technique.
• Constructed dictionaries from deep features and computed sparse approximations of Regions Of Interest (ROIs).
• Introduced block and patch decomposition methods for dictionary construction and sparse coding.

Main Results:

• Deep-feature-generated dictionaries yielded more discriminative sparse approximations than traditional methods.
• The proposed technique achieved competitive performance against state-of-the-art methods.
• Block and patch decomposition strategies simplified sparse coding and found tractable solutions.

Conclusions:

• Deep feature-based sparse approximation is effective for classifying breast masses.
• The developed technique offers a promising approach for computer-aided diagnosis of breast cancer.
• This method has the potential to improve diagnostic accuracy and reduce mortality rates.