A 3D reconstruction method based on multi-views of contours segmented with CNN-transformer for long bones.

Abstract

In computer-assisted diagnosis for orthopedic treatment, 3D reconstruction of bones is critical. Traditional 3D imaging technologies like CT and MRI have been proposed, but their high radiation dose and the requirements for lying postures could impact the accuracy of reconstructed bones and diagnosis results. Meanwhile, methods based on bone contours always depend on prior knowledge and lack precise bone segmentation methods. To address these issues, a bone reconstruction method based on multi-views of contours is proposed, as well as a hybrid CNN-Transformer approach for bone contours segmentation. A four-step strategy is introduced including segmenting bone contours from X-ray images, calculating 3D sparse, dense point clouds based on contours, and reconstructing surface. The Trans-DetSeg approach for interest regions detection and bone segmentation is proposed for accurate contours. Besides, the mathematical description of mapping relationships between contours in different views of X-ray images is provided. Then, bone sparse and dense point clouds are generated subsequently. Based on dense point clouds and the power crust method, realistic bone models are reconstructed. Evaluated on 301 bone X-ray images and by considering p-value < 0.05, the proposed Trans-Detseg approach performed better with Dice Similarity Coefficient of 0.949 and Hausdorff Distance of 26.17 than three state-of-the-art models. Furthermore, the accuracy of the bone 3D reconstruction was investigated in three tibia cases and the proposed method was verified based on comparisons of results and CT data. It was proved that increased views of X-ray images could reduce the Average Surface Distance and perfect the structure information of reconstructed bones. A new method for bone 3D reconstruction based on segmented bone contours on multi-views of X-ray images has been developed. Besides, a hybrid CNN-Transformer approach is introduced to segment bone contours. Evaluations proved the efficiency and accuracy of the proposed bone 3D reconstruction method.