A Cloud Collaborative Healthcare Platform Based on Deep Learning in The Segmentation of Maxillary Sinus

Abstract

This paper proposes a novel approach called ADLM++ using deep learning and active learning methods to tackle the complex training data process and achieve maxillary sinus segmentation. Prior to dental implant surgery, dentists spend lots of time diagnosing medical images due to the large variations in the shape, size, and position of the maxillary sinus among patients. The proposed method can reduce dentists' unnecessary workload and alleviate developers' required effort to prepare training data. The model can be trained with less training dataset through active learning methods and achieve better segmentation results. Then, the final training model is built as a pre-trained model and deployed in the proposed cloud collaborative healthcare platform. After dentists upload medical images, the platform automatically extracts the maxillary sinus and converts it into a 3D model, along with its volume and surface area. This information provides dentists with a visual virtual patient for accurate treatment. In the experiment, we compared the method of manual annotation by dentists with the method of segmentation using deep learning. The proposed method improves efficiency by five times and achieves a Dice Similarity Coefficient (DSC) evaluation score of 0.971±0.003. Additionally, it can improve the precision of diagnosis and surgical planning for dentists and alleviate the problem of relying on experience for diagnosis.