Pairwise FCM based feature weighting for improved classification of vertebral column disorders

Area of Science:

• Medical imaging analysis
• Machine learning for healthcare
• Data science in diagnostics

Background:

• Vertebral column disorders such as disk hernia (DH) and spondylolisthesis (SL) require accurate classification for effective treatment.
• Existing classification methods may face challenges with complex medical datasets, necessitating improved pre-processing techniques.

Purpose of the Study:

• To propose an innovative data pre-processing method to enhance the classification performance of vertebral column disorders.
• To automatically classify disk hernia (DH), spondylolisthesis (SL), and normal (NO) groups within a vertebral column dataset.

Main Methods:

• A pairwise fuzzy C-means (FCM) based feature weighting method was developed for a three-class dataset (DH-SL, DH-NO, SL-NO).
• The method involved grouping the dataset pairwise, applying FCM for feature weighting, and subsequently classifying these weighted groups using MLP, k-NN, Naive Bayes, and SVM algorithms.
• Overall performance was evaluated by averaging accuracies from pairwise classifiers, using metrics like accuracy, sensitivity, specificity, ROC curves, and f-measure.

Main Results:

• Without feature weighting, f-measure values ranged from 0.7021 to 0.7738 across different classifiers.
• With the proposed pairwise FCM feature weighting, f-measure values significantly increased to a range of 0.9313 to 0.9603.
• The pairwise FCM method demonstrated robust and effective performance improvements for all tested classifiers.

Conclusions:

• The proposed pairwise fuzzy C-means based feature weighting method is effective for improving vertebral column disorder classification.
• This innovative approach shows promise for application to more complex medical datasets with a higher number of classes in the future.