Conserved Self Pattern Recognition Algorithm with NovelDetection Strategy Applied to Breast Cancer Diagnosis

摘要

This paper presents a novel approach based on an improved Conserved Self Pattern Recognition Algorithm to analyze cytologicalcharacteristics of breast fine-needle aspirates (FNAs) for clinical breast cancer diagnosis. A novel detection strategy by couplingdomain knowledge and randomized methods is proposed to resolve conflicts on anomaly detection between two types of detectorsinvestigated in our earlier work on Conserved Self Pattern Recognition Algorithm (CSPRA). The improved CSPRA is appliedto detect the malignant cases using clinical breast cancer data collected by Dr. Wolberg (1990), and the results are evaluated forperformance measure (detection rate and false alarm rate). Results show that our approach has promising performance on breastcancer diagnosis and great potential in the area of clinical diagnosis. Effects of parameters setting in the CSPRA are discussed, andthe experimental results are compared with the previous works.