Development of Severity Index for Oral Epithelial Dysplasia Using Fuzzy Group Decision Making Algorithm

Background Oral epithelial dysplasia (OED) grading suffers from several levels of uncertainties and imprecision. An index is preferred in this context to improve reliability and reduce subjectivity in diagnostic decision making. In this study, a fuzzy logic-based disease severity index (SI) is formulated considering standard histopathological features used in OED grading. Methods Oral onco-pathologists were asked to independently provide weights of different features, according to their clinical significance in the context of dysplasia. Aggregated weight of each feature was calculated from individual assessment of different onco-pathologists by a fuzzy logic-based group decision making algorithm. Confidence levels of experts were also included to improve robustness of the method. SI was generated by integrating abnormality score of each feature with its weight. Abnormality degree of each feature was expressed in linguistic terms by onco-pathologists which were subsequently represented by a triangular fuzzy number. Fuzzy membership function was used as it can capture the ambiguity of experts’ opinion regarding abnormality of individual feature. Finally, defuzzification was used to get a crisp index from weighted sum of all features. Result SI was found to be statistically different (p<0.01) for different grades of dysplasia i.e mild, moderate and severe with added advantage of stratifying each grade in low and high subcategory. Conclusion Key contribution of our work is that we have developed a fuzzy logic-based group consensus process regarding weights of histopathological features for OED grading. Present methodology of developing SI can be applied to other medical decision-making problems as well. Highlights A severity index is proposed to improve reproducibility of oral epithelial dysplasia grading. Clinical significance of each histopathological and cytopathological feature in the context of dysplasia is reflected by its weight. A fuzzy logic-based group decision making algorithm is used to reduce subjectivity of features’ weights. Confidence level of oral onco-pathologists are given due importance in derivation of aggregated weight of each feature. Abnormality score of each feature is evaluated in fuzzy scale to capture clinicians’ ambiguity. Present methodology may be useful for developing indices of other diseases. Abstract Figure Graphical abstract