On the use of multi-objective evolutionary classifiers for breast cancer detection

Area of Science:

• Medical imaging
• Computational biology
• Machine learning

Background:

• Breast cancer is a common women's cancer, making early detection via mammography crucial.
• Routine mammograms are vital for early breast cancer detection and treatment.

Purpose of the Study:

• To describe, analyze, compare, and evaluate three image descriptors for breast cancer image classification.
• To assess the performance of Multi-Objective Evolutionary Algorithms (MOEAs) in classifying breast cancer images.

Main Methods:

• Utilized four medical image databases with reliable human annotations.
• Applied and compared various combinations of classification objectives within MOEAs.
• Conducted empirical analysis supported by statistical approaches to evaluate MOEA performance.

Main Results:

• Statistical tests indicated no single superior algorithm for comparing diverse techniques or evolutionary algorithms.
• Evaluating meta-classifiers with a single metric can skew results and is inadvisable.
• MOEA performance was measured against gold standard classifications.

Conclusions:

• Selecting the appropriate set of objectives and criteria is essential for robust classification.
• Accuracy-related objectives are directly linked to maximizing true positives.
• Relying solely on generic accuracy metrics can shift the primary classification goal inappropriately.