Spatial Field Fusion Network (SFFNet) for Panoramic Dental X-ray Segmentation

Deep learning models in the medical industry have been in high demand as they minimize the time and effort required for manual diagnosis. We introduce Spatial Field Fusion Network (SFFNet), a deep neural network as a computeraided diagnosis (CAD) based solution for dental segmentation in panoramic X-rays. SFFNet is examined on the UFBA UESC Dental Images Deep Dataset, which consists of 1,500 images of 10 distinct categories. SFFNet performs better than state-of-the-art baseline networks on the test set of 1225 images after training on only 193 images, obtaining 95.46% dice-coefficient, 96.53% precision, 95.76% recall, and 96.11% Fl-score. SFFNet is a lightweight architecture with model parameters as little as 0.2 million, making it a preferred choice for edge devices.