Wang, Wei
(2020)
Intelligent Planning for Refractive Surgeries: A Modelling and Visualisation-based Approach.
PhD thesis, University of Liverpool.
|
Text
201152300_June2020.pdf - Unspecified Download (6MB) | Preview |
Abstract
Laser refractive surgeries have been commonly used in ophthalmic operations. Considerable research has been carried out and encouraging progress made in recent years. It covers properties of the cornea and behaviour of tissue in different parts of the eye, topography and material expression of individual patient's eyes, prediction using finite element (FE) analysis to estimate the corneal shape change and the change in refractive power. Further effort is still required to advance the research to aid the decision making for laser refractive surgeries. This study comprehensively reviews the latest techniques of refractive surgery and research on computational analysis and modelling techniques and their applications, especially the current prediction and planning techniques for laser refractive surgeries. The aim of this study is to develop an intelligent assistant tool for the laser refractive surgeries with prediction and visualisation functions. For this aim, two objectives will be achieved: prediction with the clinical dataset and human vision simulation. Due to clinical statistics, the clinical dataset is often incomplete, imbalanced, and sparse. Three methods are proposed to predict surgery parameters and outcomes using the clinical dataset. A multiple imputation method, with multiple regression, is proposed for imputing the missing data. For the imbalance of data distribution in the clinical dataset, an over-sampling of the minority data method is proposed. The accuracy of predicted minority data is close to the accuracy of predicted majority data. Finally an ensemble learning method which is optimised by the genetic algorithm is proposed to improve the accuracy of the prediction results with a sparse dataset. According to the distribution of the sample in the clinical data, the percentage of unacceptable results is 23.02%. The methods in this study could provide an accuracy of 79.02% to find the possible unacceptable cases, that is, the method could reduce the percentage of unacceptable results from 23.02% to 4.82%. In human vision simulation, the study focuses on how the human vision simulation could be determined and obtained accurately within a required timeframe. The ray tracing technique can provide more precise results than the rasterisation technique, especially for the simulation of light reflection and refraction in the human eyeball. However, the thin lens assumption affects the accuracy of the pathological vision simulation with the ray tracing technique. An improved schematic human eye model is proposed to obtain a numerical model predicting the size of the defocus blur for the pathological vision, which wraps the shape of the ray intersection area. In order to generalise this model to other healthy and pathological vision, an intelligent blur range derivation method is proposed. On the other hand, ray tracing scene rendering requires repeated iterative computing which takes a significant amount of computation time. A GPU-based ray tracing computing method is proposed to accelerate and optimise the rendering of scenes. With this method, the scene rendering speed is about 75 times faster than using the CPU.
| Item Type: | Thesis (PhD) |
|---|---|
| Divisions: | Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science |
| Depositing User: | Symplectic Admin |
| Date Deposited: | 14 Aug 2020 09:21 |
| Last Modified: | 18 Jan 2023 23:49 |
| DOI: | 10.17638/03090577 |
| Supervisors: |
|
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3090577 |
Dimensions
Dimensions
