Otitis media detection using tympanic membrane images with a novel multi-class machine learning algorithm.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Otolaryngology

Background:

• Otitis media (OM) is a common middle ear infection diagnosed via subjective otoscope image inspection.
• Current diagnostic methods are prone to errors and subjectivity.
• There is a need for objective and reliable diagnostic tools for OM.

Purpose of the Study:

• To develop a novel computer-aided decision support model for diagnosing otitis media.
• To enhance the model's generalization ability using advanced deep learning techniques.
• To provide an objective and repeatable diagnostic aid for otitis media.

Main Methods:

• A convolutional neural network (CNN) based model was developed.
• The model incorporates channel and spatial attention (CBAM), residual blocks, and hypercolumn techniques.
• Experiments were conducted on an open-access dataset of 956 otoscope images across five classes.

Main Results:

• The proposed CNN model achieved high classification performance.
• Overall accuracy was 98.26%, with 97.68% sensitivity and 99.30% specificity.
• The model outperformed pre-trained CNNs like AlexNet, VGG-Nets, GoogLeNet, and ResNets.

Conclusions:

• The developed CNN model with integrated image processing techniques is effective for otitis media diagnosis.
• This AI tool can assist specialists in achieving objective, repeatable results and reducing misdiagnosis.
• The model supports clinical decision-making, improving the accuracy and efficiency of OM diagnosis.