A Maximum-Entropy Fuzzy Clustering Approach for Cancer Detection When Data Are Uncertain

(1) Background: Cancer is a leading cause of death worldwide and each year, approximately 400,000 children develop cancer. Early detection of cancer greatly increases the chances for successful treatment, while screening aims to identify individuals with findings suggestive of specific cancer or pre-cancer before they have developed symptoms. Precise detection, however, often mainly relies on human experience and this could suffer from human error and error with a visual inspection. (2) Methods: The research of statistical approaches to analyze the complex structure of data is increasing. In this work, an entropy-based fuzzy clustering technique for interval-valued data (EFC-ID) for cancer detection is suggested. (3) Results: The application on the Breast dataset shows that EFC-ID performs better than the conventional FKM in terms of AUC value (EFC-ID = 0.96, FKM = 0.88), sensitivity (EFC-ID = 0.90, FKM = 0.64), and specificity (EFC-ID = 0.93, FKM = 0.92). Furthermore, the application on the Multiple Myeloma data shows that EFC-ID performs better than the conventional FKM in terms of Chi-squared (EFC-ID = 91.64, FKM = 88.26), Accuracy rate (EFC-ID = 0.71, FKM = 0.60), and Adjusted Rand Index (EFC-ID = 0.33, FKM = 0.21). (4) Conclusions: In all cases, the proposed approach has shown good performance in identifying the natural partition and the advantages of the use of EFC-ID have been detailed illustrated.