Automated breast image classification using features from its discrete cosine transform.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Breast cancer screening programs aim for high accuracy to detect malignancies early.
• Automated classification methods offer potential to enhance diagnostic performance and reduce errors.
• Minimizing false negatives is critical in breast cancer detection to ensure timely treatment.

Purpose of the Study:

• To improve breast screening accuracy through automated classification using whole image features.
• To evaluate the efficacy of discrete cosine transform (DCT) features for classifying breast tissue.
• To prioritize the reduction of false negative findings in breast cancer detection.

Main Methods:

• Utilized online datasets for developing and testing classification algorithms.
• Implemented image processing using Mathematica and C++ for computational efficiency.
• Applied discrete cosine transform (DCT) to extract features from breast tissue images.
• Tested k-nearest neighbor and naive Bayesian classifiers on DCT-derived features.

Main Results:

• Generated and tested 41 features derived from DCT.
• Achieved sensitivities up to 98% with 66% specificity using k-nearest neighbor.
• Attained sensitivities up to 100% with 64% specificity using naive Bayesian classifiers.

Conclusions:

• Discrete cosine transform (DCT) features effectively support whole image classification for breast cancer screening.
• The high sensitivity achieved suggests DCT features can augment existing screening methods to increase overall accuracy.
• Classifiers with near 100% sensitivity, like the one developed, can be integrated into sequential analysis to enhance diagnostic performance.