Fully Automatic Segmentation of Sphenoid Sinus in CT Images with 3D Convolutional Neural Networks

Kamal Souadih, Ahror Belaid, Douraied Ben Salem

Abstract:

Today, Deep learning algorithms have quickly become essential in the field of medical image analysis. Compared to the traditional methods, these Deep learning techniques are more efficient in extracting compact information leading towards significant improvement performance of medical image analysis system. We present in this paper a new technique for Sphenoid sinus automatic segmentation using a 3D Convolutional Neural Networks (CNN). Due to the scarcity of medical data, we chose to used a 3D CNN model learned on a small training set. Mathematical morphology operations are then used to automatically detect and segment the region of interest. Our proposed method is tested and compared with a semi-automatic method and manual delineations made by a specialist. The preliminary results from the Head Computed Tomography (CT) volumes
seem to be very promising.

Keywords: 3D Convolutional neural network, Computed Tomography, Automatic Segmentation, Sphenoid Sinus