- 學位論文
頻域式光學同調斷層之影像處理於高度近視患者之青光眼診斷的應用
The Application of Image Processing for Fourier Domain Optical Coherence Tomography Image in Glaucoma diagnosis for High Myopia Patient
摘要
青光眼為眼科疾病中常見之致盲疾病之一。所有治療方式僅能暫緩視野惡化程度，因此青光眼疾病早期發現、早期治療是相當重要的。經頻域式光學同調斷層掃描(Fourier Domain Optical Coherence Tomography, FD-OCT) 找出神經節細胞複合層(GCC)之厚度為為目前早期偵測青光眼最重要之指標。對於高度近視患者，由於眼軸拉長進而導致視網膜被拉長而變薄，造成GCC厚度判斷上的錯誤，干擾醫師青光眼的診斷。因此，本研究研發一近視校正模型，藉由數學計算及影像處理技術，用於FD-OCT所得到的黃斑部視網膜斷層掃描影像，來修正視網膜厚度數據，藉此資訊輔助醫師，而提高了高度近視患者在診斷青光眼時的準確率。 本研究採用FD-OCT於黃斑部上之斷層掃描，並利用插值法模擬FD-OCT掃描線之外的點，得到完整之黃斑部GCC厚度資料，並使用近視校正模型校正GCC厚度，得到兩個特徵參數：平均GCC厚度、GCC厚度上下半部不對稱性 (IAsymmetric) 。再分別引用閥值與類神經網路的兩種分類法進行分類。使用材料為經由醫師診斷結果之青光眼病患18位，左右眼共36個，其中包含早期青光眼18個(高度近視12個、非高度近視6個)，重度青光眼18個(高度近視8個、非高度近視10個)；正常人15個，左右眼共30個(高度近視12個、非高度近視18個)之FD-OCT影像。系統評估方式分別引用閥值與類神經網路的兩種分類法，將材料分為測試組與訓練組，對分辨正常人與青光眼(包含早期與重度)，進行訓練與測試。 初步結果顯示校正後對於分類正常人與青光眼確能改善其鑑別能力。在閥值分類法在校正前後其系統評估參數，準確率由0.7提升至0.91、靈敏度維持在1、專一性由0.33提升至0.8、Kappa 值由0.35提升至0.81；在類神經網路分類法在校正前後其系統評估參數，準確率由0.88提升至0.97、靈敏度由0.78提升至0.94、專一性由維持在1、Kappa值由0.76提升至0.94。此外，若單就正常人與早期青光眼患者分類來看，近視校正法其降低誤判率之能力更為顯著。 本研究發展了以黃斑部GCC厚度近視校正模型及兩個特徵參數(平均GCC厚度與上下半部GCC厚度上下半部不對稱性)為基礎之早期青光眼檢測系統。使用此系統可有效的分出正常人與早期青光眼患者，但對於早期青光眼與重度青光眼之分辨能力則未達預期。
並列摘要
Glaucoma is one of familiar blind diseases in the ophthalmology. Glaucoma only can temporarily decelerate worse degree after treatments, thus, it’s important to discover and treat this disease in early stages. The Ganglion Cell Complex (GCC) thickness which can be detected by Fourier Domain Optical Coherence Tomography (FD-OCT) is the most important indicator for early glaucoma patient. For a patient with high myopia, the axial length of eyeball will be stretched, furthermore as the retinal have been extent as thinning. It may cause the calculation of the thickness of GCC has be failing, and interfere to the diagnosis of glaucoma. Therefore, in order to increase the accuracy in diagnosis the glaucoma for high myopia patient, a myopic correction model based on mathematical calculations and image processing technology was developed in this study, to correct the thickness in macular retinal tomography images for FD-OCT. The FD-OCT scans of macular, and simulating the point out of scans by Cubic Spline Interpolation was used to complete full data of GCC thickness of macular. And then two characteristic parameters, mean GCC thickness and index of asymmetric, were obtained by using myopic correction model to modify the GCC thickness. Finally, two kind of classification method, classifying by threshold and classifying by neuron network, were used. This procedure was applied to 18 patients (36 eyes) with glaucoma and 15 normal subjects (30 eyes) which diagnosed by physician. In 36 eyes with glaucoma include 18 eyes with early glaucoma (12 eyes with high myopia and 6 eyes with myopia less -6D) and 18 eyes with advance glaucoma (8 eyes with high myopia and 10 eyes with myopia less -6D). In 30 eyes of normal subjects include 12 eyes with high myopia and 18 eyes with myopia less -6D. In system evaluation, the material was divided into training group and the test group, to classify normal subjects and glaucoma (include early glaucoma and advance glaucoma). After the myopia modification and in classifying by threshold method, beside sensitivity remained at 1, accuracy, specificity, and kappa value increased from 0.7 to 0.91, from 0.33 to 0.8, and from 0.35 to 0.81, respectively; And in classifying by neuron network method, beside specificity remained at 1, accuracy, sensitivity, and kappa value increased from 0.88 to 0.97, 0.78 to 0.94, and upgrade from 0.76 to 0.94, respectively. Results show that the ability can be enhancing of classification of normal subjects and glaucoma by myopia modification. Moreover, the effect of myopia modification for classification of normal subject and early glaucoma shows much better performance. A system based on the myopic correction model for modification GCC thickness of macular and two characteristic parameters (mean GCC thickness and index of asymmetric) was developed in this study. By using this system, it is easier to distinguish normal and early glaucoma patients. However, the ability to classify between early glaucoma and advance glaucoma was not significant as expected.