利用空間相關閾值做深度學習分類以偵測黃斑部  病變之光學同調斷層掃描影像

視網膜黃斑部病變為老年或糖尿病患者的常見症狀。其症狀有視線內盲點，視點歪斜，嚴重可導致失明。醫學上多以光學同調斷層掃描影像，來判斷病人是否為黃斑部病變的患者。在斷層掃描影像上，黃斑部會有積水或是增生組織的產生，需以專業醫生以人工的方式來判斷病變的範圍。本研究利用機器學習取代傳統人眼辨識。深度學習也稱人工智慧，對於視網膜黃斑部病變範圍訓練一個專屬神經網路。其輸入為病人的視網膜斷層掃描影像，輸出為病變的範圍，實現了自動辨別病變範圍的成果。其訓練過程主要分4個步驟一.利用空間相關閾值得到感興趣區域。二.使用感興趣區域訓練新的ROI判別模型。並且根據感興趣區域遮罩切割原圖。三.將切割好的視網膜圖片進行訓練。並且和真實病變範圍比較。重複訓練神經網路。四.訓練好的神經網路，輸入未曾看過的圖片，也能得到正確的病變範圍。實現了自動辨別病變範圍的成果。本研究使用了新的方式，利用模糊矩陣和空間可調式閾值，取代傳統邊界演算法，達到取特定區域的效果。其特色為快速，簡單，且不需訓練。其結果和先前的論文幾近相同。在特殊難以判別的區域，本論文也提供深度學習法。利用神經網路強大的特徵判斷力，製作出能夠根據視網膜圖片選取特定區域的模型。經過此模型的遮罩，切割出視網膜圖片的特定區域，再進行黃斑部水腫病變的神經網路，執行病變範圍的判斷。空間相關閾值的方式也可利用在各種影像處理上，成為任何影像處理的第一步驟。在沒有醫生人工標籤的情況下，依然能夠使用影像處理方式產生人工標籤，藉此訓練感興趣區域的神經網路。提升黃斑部病變神經網路的辨別能力。

關鍵字

深度學習；黃斑部水腫；感興趣區域

並列摘要

Macular degeneration is a common disease occurs in elder or people who have diabetes. Optical coherence tomography (OCT) has been used to diagnosis the Age macular degeneration (AMD), and diabetic macular edema (DME). The fluid regions in retina are the most characteristic of AMD, and it could be observed in retinal OCT imaging by a ophthalmologist. We propose an automatic machine learning method to segment the AMD regions. There are four steps to accomplish the segmentation. 1.Using the spatial adaptive threshold to segment the region of interest (ROI). 2.Cropping the OCT image by the ROI mask. 3.Training the neural network and modified by the manual AMD ground truth. 4.Inputting the raw data to the neural network, and obtain a automatic segmentation of the AMD regions. In the paper, we used a blurring convolution metric and spatial adaptive threshold to obtain the ROI, instead of the previous works using deep learning method, canny, or other edge detector. Spatial adaptive threshold is a simple, fast way comparing with the previous work, and it has same quality. It can also be the first step of all the image processing.

並列關鍵字

deep learning ； ROI ； Segmentation

參考文獻

1. A. Rashno, D. Koozekanani, P.M. Drayna, B. Nazari, S. Sadri, H. Rabbani and K.K. Parhi. “Fully-automated segmentation of fluid/cyst regions in optical coherence tomography images with diabetic macular edema using neutrosophic sets and graph algorithms,” IEEE Transactions on Biomedical Engineering, Issue 5, pp.989- 1001. (2017)

Google Scholar

2. 4 A2A SD-OCT clinics (Devers eye institute, duke eye center, emory eye center, and national eye Institute)

Google Scholar

3. L. Fang, D. Cunefare, C. Wang, R.H. Guymer, S. Li, and S. Farsiu. “Automatic segmentation of nine retinal layer boundaries in OCT images of non-exudative AMD patients using deep learning and graph search,” OSA Biomedical Optics Express, vol.8, Issue 5, pp.2732-2744. (2017)

Google Scholar

4. J. Long, E. Shelhamer, T. Darrell, UC Berkeley. “Fully convolutional networks for semantic segmentation,” arXiv Computer Science, 1411.4038[cs.CV]. (2015)

Google Scholar

5. S. Albawi, T.A. Mohammed and S. Al-Zawi, “Understanding of a convolutional neural network,” IEEE, 2017 International Conference on Engineering and Technology (ICET), Antalya, 2017, pp.1-6. (2017)

Google Scholar

國際替代計量

利用空間相關閾值做深度學習分類以偵測黃斑部病變之光學同調斷層掃描影像

全文下載

主題瀏覽