Adaptive neuro-fuzzy inference system for classification of ACL-ruptured knees using arthrometric data.

Abstract

A new approach, based on Adaptive-Network-based Fuzzy Inference System (ANFIS), is presented for the classification of arthrometric data of normal/ACL-ruptured knees, considering the insufficiency of existing criteria. An ANFIS classifier was developed and tested on a total of 4800 arthrometric data points collected from 40 normal and 40 injured subjects. The system consisted of 5 layers and 8 rules, based on the results of subtractive data clustering, and trained using the hybrid algorithm method. The performance of the system was evaluated in four runs, in the framework of a 4-fold cross validation algorithm. The results indicated a definite correct diagnosis for typical injured and normal cases. Except for two, all cases with marginally distinct force-displacement curves were also diagnosed correctly. The overall sensitivity and specificity of the system in four runs were 95.5% and 100%, respectively. The superior performance of the ANFIS classifier over previously suggested criteria highlights its capability when dealing with marginal arthrometric data of knees with partially disrupted ACL or hypermobility syndrome.