Hybrid Intelligent System for Otitis Media Detection Using YOLOv12 and Fuzzy Logic

Otitis media (OM) is a middle-ear condition that can affect children and adults; delayed recognition may lead to hearing-related complications.Interpreting otoscopic images depends on the clinician's assessment of the tympanic membrane and can vary with experience.This study presents a hybrid image-analysis pipeline that combines YOLOv12 for tympanic-membrane localization, a convolutional neural network (CNN) for five-class classification, and a Mamdani fuzzy-inference stage intended to make the final output more interpretable.The classes are Normal, Acute Otitis Media, Chronic Otitis Media, Myringosclerosis, and Cerumen Impaction.The fuzzy stage uses redness, opacity, bulging and fluid-related scores, together with detector confidence, in a 12rule system.For the 3,000-image classification dataset, the CNN obtained 93.17% accuracy and a macro F1score of 0.9323.The YOLOv12 detector reported a box mAP@0.50 of 0.995 on the detection evaluation.The fuzzy stage was assessed on a 50-image held-out sample; its agreement with the reference labels was low, so its rules and membership functions require further calibration.These findings describe a prototype for further evaluation, not a clinically validated diagnostic system.