Study On Sella Turcica Morphology Based On Deep Learnin

The morphology of sella turcica can help doctors predict the growth and development of patients,evaluate craniofacial morphological characteristics,and find abnormalities of pituitary gland.More and more studies have focused on the morphological changes of sella turcica,hoping to use the morphology of sella turcica as a reference for clinical diagnosis.So far,the linear parameters of sella turcica can only be obtained by manual measurement.However,manual measurement is not only time-consuming and laborious,but also prone to subjective errors.Therefore,this thesis develops and evaluates a method based on deep learning for automatic measurement of sella turcica morphology.The research contents are divided into two parts:(1)Automatic segmentation of sella turcica.A dataset was constructed,which included X-ray images and the masks generated by annotation.The U-net model was trained to realize automatic segmentation of sella turcica.The images are preprocessed before segmentation,and the segmentation results are postprocessed after segmentation to improve the segmentation accuracy of the model.The segmentation performance of the model was evaluated using the Dice coefficient,and the result reached 0.9284.The trained segmentation model is used to complete the automatic segmentation of all images in the test dataset,and the segmentation results are used in the second part of the research.(2)Automatic measurement of linear parameters.Firstly,the results of automatic segmentation are processed by the extremum points detection algorithm and the corner points detection algorithm,so that the four landmarks on the sella turcica contour are automatically located and their coordinates are returned.Secondly,the coordinates are calculated using the distance formula between two points and the distance formula from point to straight line to obtain the length,diameter,and depth of the sella turcica.In addition,the linear parameters of the sella turcica are measured manually by Digimizer.Finally,the intraclass correlation coefficient and Bland-Altman plots were used to analyze the consistency between the automatic measurement results and manual measurement results of the linear parameters of sella turcica.The analysis results showed that the automatic measurement method proposed in this thesis has good reliability.This thesis explores and realizes the automatic measurement of sella turcica morphology for the first time.The proposed automatic measurement method can promote the further development of the study on the morphological of sella turcica.