Acute Otitis Media (AOM) is a common middle ear infection, especially in children, but it can happen at any age. Accurate and early detection is essential for successful treatment and avoiding complications. This research introduces a new method for AOM detection through an Improved Mask R-CNN (IR-CNN) model combined with ResNet-50. The method adopted here is systematic and starts with the preprocessing of the images, comprising Wiener filtering for denoising and K-means clustering in the Lab color space for segmentation. The feature extraction step is carried out by convolution layers in the IR-CNN, followed by the classification process done by a three-layer Fully Connected Neural Network (FCNN) to separate between normal and AOM-affected cases. Performance of the system is measured based on accuracy, sensitivity, specificity, F1-score, AUC, and IoU metrics. K-Fold Cross Validation has been used to measure the strength and generalizability of the model. The system is very effective in the diagnosis of AOM with high accuracy of 99.1%, precision of 98.99%, and F-score of 98.45%, highlighting its efficacy for enhancing clinical decision-making.