Volume Estimation of Renal Stone on Computed Tomography Imaging Using Deep Learning
| Autor: | YANG, CHENG-YEN, 楊承諺 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 Kidney stones can affect the urinary tract system in human body. While kidney stones are passing through the ureter it can be quite painful and sometimes become lodged in the urinary tract. Clinically to diagnose kidney stones, it is commonly to use computer tomography (CT) scan with physicians making assessment manually. To reduce human workload, many approaches has been proposed in the literature. Recently, semantic segmentation tasks in color images are improved via deep learning model. Especially, strategies for cell image segmentation show a good success. Therefore, a system for kidney segmentation using deep learning and model reconstruction is proposed. The system consists of three steps, image preprocessing, region of interest (ROI) selection and model reconstruction. Firstly, edge detection is applied to locate the spine in the image. Since the position of kidney is usually between (T12) spine and (L3) spine, ROI can be generated between them. After that a pixel-based convolutional neural network is developed to segment the kidney contours from CT. The resulting masks is then used to reconstruct the kidney model. Besides, kidney stones are detected from the region with their shape, size and position measured. In this thesis, 30 data-sets from MICCAI 2019 Kidney Disease Analysis Challenge and 33 data-sets provided by the Keelung Government Hospital were used for experiments. According to Dice coefficient, the proposed method can achieve 94.8% and 98.3% similarity in kidney segmentation and in kidney stones segmentation, respectively, compared to groundtruth. The proposed system has demonstrated its feasibility in assisting diagnoses, which can further achieve the goal for reducing medical cost and improving healthcare quality. Keywords: automation, image processing, contour extraction, volume estimation |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|