Research On Calibration Method Of Eye Movement Data Acquisition Equipment

Objective: This study aimed to investigate the accurate segmentation of the clinical target volume of cervical cancer based on preoperative computed tomography(CT)images of 214 patients with T1,T2,T3,and T4 a stage cervical cancer.The spatial information of the anatomical position,shape,and size of the target area was obtained,and a model framework based on end-to-end deep learning was proposed to automatically segment the clinical target volume of cervical cancer,providing reference for doctors to formulate plans for radiation therapy and surgical procedures.Methods: Based on a retrospective collection of preoperative CT images,radiation therapy plans,and radiation dose data from 214 cervical cancer patients and a comprehensive study of the characteristics of cervical cancer lesions in CT images,a threestage workflow was proposed to achieve the end-to-end process from preoperative CT images to outputting the clinical target volume of cervical cancer.In the first stage,a slice detection model was used to classify CT images to select continuous CT slice images containing the clinical target volume.In the second stage,a dose contour feature map was generated using the radiation dose data and the selected CT slices.In the third stage,the dose contour feature map was fused with the CT slices and input into the encoder-decoder model to achieve accurate segmentation of the clinical target volume of cervical cancer.Results: In the test set containing 1796 T1,T2,T3,and T4 a stage CT image slices from42 patients,the proposed three-stage segmentation model based on the encoder-decoder structure improved the deficiency of unclear edge contours in segmenting the clinical target volume of cervical cancer.The Dice coefficient,MIOU coefficient,and PA coefficient were 95.35±0.14%,90.54±0.27%,and 99.37±0.03%,respectively,in the three objective evaluation indicators.Through three ablation experiments,it was found that our model can improve the accuracy of the clinical target volume segmentation of cervical cancer.Conclusion:The proposed method can select slices containing the clinical target volume of cervical cancer without missing any,and can improve the deficiency of unclear edge segmentation in the existing automatic segmentation model for cervical cancer radiation therapy by extracting the radiation dose contour feature map to assist in correcting the target volume edge contour.The method provides more accurate reference information for doctors to implement radiation therapy and surgery and has important clinical application value.