Sinusitis Diagnosis in Temporal Bone CT Using Extremal Intensity Transform-Based Inflated 3-D ConvNet

Despite being primarily used for ear assessment, temporal bone CT (TBCT) can be an effective diagnostic tool for sinusitis because it captures parts of paranasal sinuses, especially given the strong link between ear diseases and sinusitis. In this work, we propose a TBCT-based sinusitis diagnosis method using deep learning. As a pioneering research problem, the available training data for this is limited besides the high dimensionality of 3D TBCT volumes, causing overfitting. For improved training, we propose an extremal intensity transform to simplify the model input and incorporate an inflated 3D convolutional network for effective pre-training. We evaluated our diagnostic method on a private dataset of 100 TBCT volumes collected from a local hospital. The proposed approach achieved an average area under the receiver operating characteristic curve (AUC) of 0.95 and an accuracy of 0.90, significantly reducing overfitting while enhancing diagnostic accuracy by 5% compared to standard deep learning-based diagnostic models.