Fuzzy classification of patient state with application to electrodiagnosis of peripheral polyneuropathy

A methodology which accounts for uncertainty or imprecision in experimental observations and both norm and pathology definitions is developed on the basis of a distance measure between fuzzy numbers. These fuzzy numbers may represent, respectively, the measurements, norm, and pathology. The distance measure, called normalized fuzzy pathology index (NFPI), evaluates the difference of distance between observed experimental values for a given patient and norm on the one hand, and pathology on the other hand. The NFPI characterizes patient state as a continuous index; however, to conform to medical usage, categories of values are defined. Each of these categories corresponds to a linguistic variable. The case study used to illustrate the methodology is the electrodiagnosis of peripheral polyneuropathy in diabetic patients. Here, four initial linguistic categories are defined by a physician, namely: normal state, borderline state, clear-cut, and severe pathology. The NFPI is calculated in three cases that provide a sensitivity analysis on measurement fuzziness and distance function weighting. The model is calibrated using 203 cases and validated using 291 different cases. The results correspond very closely to the physician's diagnosis. The loss of information in discretizing the continuous state of patients is discussed. Transferring this fuzzy approach to other cases where the concept of distance is relevant offers no difficulty.<<ETX>>