Automated AI-Driven Detection of Mastoiditis in Temporal Bone via CT Scan and MRI Imaging: A Review

Recent advances in using artificial intelligence (AI) in medical imaging toward the detection of mastoiditis have shown very promising applications. Therefore, the comprehensive review on the state-of-art approaches related to the application of CT and MRI imaging modalities for automated mastoiditis detection is provided. We review methodologies that go from conventional machine learning approaches such as SVM and Random Forests, to more novel and advanced deep learning architectures, such as U-Net, ResNet, or ViT, to image temporal bones. The review systematically assesses particular techniques: transfer learning using pre-trained models, attention mechanisms focusing feature extraction, and multi-modal fusion approaches combining the data found in CT and MRI. Window level adjustments, skull stripping, adaptive histogram equalization are subjected to critical review regarding the pre-processing methods. Its review ranges over performance metrics across the studies; the current systems achieve sensitivity rates of 85-95% and specificity of 88-97% within the controlled environment. We discuss some of the key challenges such as limited availability of dataset, variance in imaging protocols, and the need for robust clinical validation. Recent evidence from AI-driven systems is promising in the support of the process for clinical diagnosis, especially when using ensemble approaches. There could be several architectures that would work in combination to enhance the robustness and efficacy of clinical diagnosis.