Implementing Novel Kernel Discriminant Analysis to Optimize Breast Cancer Diagnosis

Overview

Objectives/Goals The objective is to determine if pre-processing mammographic data with KDA (Kernel Discriminant Analysis) increases the accuracy of a machine-learning breast cancer diagnosis algorithm by increasing detections and removing false positives. This will result in the simultaneous advancement of the medical and machine learning fields. Methods/Materials Mammographic images of 50 normal and 50 cancerous patients were obtained from the DDSM. These files were processed, filtered, segmented, and distilled into a set of 11 points each. These points were then processed with Quadratic (QDA) and Linear Discriminant Analysis (LDA) and stored separately. Three different machine learning algorithms of varying efficiencies were set up, and trained with subsections of the data, to analyze the effectivity of the Kernel Discriminant Analysis pre-processing technique over a range of data sizes and base training algorithms. Results Among the two methods tested, quadratic and linear, Quadratic Discriminant Analysis provides the greatest accuracy boost. Trends in results were uncannily similar between the three base machine learning algorithms, and followed the same general pattern. With small-scale training data sets (10-60 subjects) , QDA provides a 60% accuracy boost over the control group, and LDA a 30% increase. Under training data sets with size over 60, QDA provides comparable accuracy as the control, but LDA deteriorates and a 15% decrease in accuracy. Conclusions/Discussion The data collected shows that quadratic discriminant analysis is a viable pre-processing step for current laboratory breast cancer diagnostics. The algorithm has a marked increase in accuracy with smaller data sets, and adds low overhead, enough to justify implementation into diagnostic programs.

Competition history

• CSEF 2014 Mathematics & Software · Entry S1404

Resources

• CSEF project summary ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2014 project