Machine Learning Classifier for Retina Multi Stages Formation/Deformation Detection

Challenge Background

Diabetic Macular Edema (DME) is considered one of the most common symptoms accompanying diabetic patients that involve vision fogginess and vision loss depending on the severity of the patient’s case. Figures have revealed that 17.3 per cent of the UAE population between the ages of 20 and 79 have been diagnosed with type 2 diabetes. People with diabetes are known to be more at risk of developing eye conditions, such as cataracts, glaucoma, or retinal vessel occlusion. Diabetic retinopathy and Diabetic Macular Edema (DME) are two of the top five eye problems affecting UAE patients, so there is a huge need to raise awareness of the serious consequences, as published by Arab Health Magazine in their interview to Dr. Saleh Saif Al Messabi, Head of the Emirates Ophthalmology Society, about the impact of diabetic eye disease on the UAE society.

In order to implement automatic analysis for OCT images to detect DME patients, we need to understand the main features that distinguish the DME effect and variations over normal OCT images. These properties have been listed below based on expert knowledge:

• Cystoid Macular Edema (CME): The OCT scan in this case reveals irregular bubbles in the retina that vary in size and location. It’s worth noting that, as illustrated in Figure 1, this is the first stage of DME illness infection.

Figure 1 OCT for Cystoid macular edema [1]

• Diffuse Retinal Thickening (DRT): The retina thickness is deformed in this scenario due to local intraretinal fluid collection in the lower retinal layers, causing severe morphological changes in the eye fundus. Figure 2 depicts the buildup of subretinal fluid in an OCT scan. Furthermore, this instance is regarded as the DME’s second stage.

Serous Retinal Detachment (SRD): As a result of the separated retinal tissues, a large deformation in the outer retina appears as a dome-like elevation in this case. The fluid collection in the macular region is the cause of this distortion. It’s the third stage of DME, and it’s a rare occurrence among patients because it’s caused by another key factor: the absence of tears.

Moreover, the dome shape deformation contains a lot of bubbles inside (Cystoids) as shown in figure 3 below.

Figure 3: OCT for Serous Retinal Detachment [3]

The Problem

Diabetic Macular Edema (DME) is considered one of the most common symptoms accompanying diabetic patients that involve vision fogginess and vision loss. The primary dilemma in DME relies upon the fact that its diagnosis involves taking Optical coherence tomography (OCT) which is a non-invasive medical imaging technique, which should be examined through specialists. However, these procedures could take a long time process and involve a high cost during the specialist consultation. Moreover, OCT diagnosis could suffer from human mistakes due to many factors like poor OCT machine quality images or human diagnosis mistakes. This project investigate the opportunity to explore the best machine learning classifier to detect the retina deformation and supporting the DME identification by analysing the best features of the disease.

Goal of the Project

Diabetic Macular Edema (DME) is one of the various eye illnesses that can affect diabetic patients, especially those with Type 2 diabetes. Diabetic Retinopathy (DR), or damage to blood vessels in the eye's retina, causes fluid accumulation in the macula, resulting in DME. As a result, there is an increase in retinal thickness within one disk diameter of the fovea center, with or without hard exudates and cysts.

The project will explore the computer vision image filters (Gray-scale, threshold, extract the biggest blob, remove outliers, and morphology) over the OCT images to remove noise and extract retina curve shape. Multiple machine learning models will be tested over the extracted features dataset, and the best accuracy result will be adopted.

With the duration of 4-weeks, this project aims to:

1- Analyze the medical data images related to DME. 2- Find the best features related to DME OCT images. 3- Investigate the best approaches to extract the selected features in (2). 4- Apply multiple machine learning and deep learning models over the extracted features to find the highest accuracy model for classifying the OCT images. 5- Compare the results with previous works and analyze the output. 6- Develop an interactive interface to classify OCT images related to DME patients with the best-selected ML classifier. 7- Generate a recommendation report.

Project Timeline

What you'll learn

1. Pre-processing the OCT image.

2. Explore the polynomial curve fitting algorithm and pattern features.

3. Implementing different Machine learning models to classify OCT images.

4. Building a dashboard to visualize and present the finding in an impactful way.

5. Provide a recommendation report on the best ML classifier for DME OCT images.