Research On Lesion Segmentation And Grading Diagnosis Of Skeletal Fluorosis X-ray Images

Skeletal fluorosis is an endemic disease caused by excessive intake of fluoride.Its main symptoms include bone and joint pain,limb deformation,dyskinesia and even paralysis.The main pathological manifestation is calcification of interosseous tissue.Skeletal fluorosis can be diagnosed according to the X-ray images.Skeletal fluorosis usually occurs in underdeveloped areas,resulting in lacking of medical resources,difficult diagnosis and complex diagnosis process.In this paper,the automatic diagnosis of skeletal fluorosis is researched for the first time,and a two-stage disease diagnosis scheme of segmentation—classification is proposed,which can realize the segmentation of lesion area in X-ray images of forearm bone and the classification of lesion degree.In this paper,a two-stage diagnosis scheme of skeletal fluorosis based on segmentation Neural Network and classification neural network is proposed,and the two stages are connected by a clever multi-source information fusion algorithm.Three independent models in the classification stage are integrated into an ensemble learning classifier,which achieves better robustness.In order to solve the problem that lesion only occupies a little area in X-ray images,the middle layer feature map is extracted and multi instance learning is carried out.Finally,the classifier corrects the result through the patient’s data information.Through the verification of collected images,the two-stage diagnosis scheme of fluorosis performances satisfactory in both two-class classification and multi-class classification.Moreover,the improvement measures bring a certain improvement to the classification persion.