Retinal Damage Detection by Image-Based Deep Learning
In the modern era, we are automating almost every possible thing that comes to our mind and Machine Learning is solving many real-world problems. One of the major impacts is in the medical field. In this blog, we will go through a deep learning approach that predicts the class of the retina damage.
Retinal optical coherence tomography (OCT) is an imaging technique used to capture high-resolution cross sections of the retinas of living patients. Approximately 30 million OCT scans are performed each year, and the analysis and interpretation of these images take up a significant amount of time (Swanson and Fujimoto, 2017).
OCT is useful in the diagnosis of many retinal conditions, especially when the media is clear. In general, lesions in the macula are easier to image than lesions in the mid and far periphery. OCT can be particularly helpful in diagnosing:
In some cases, OCT alone may yield the diagnosis (e.g. macular hole). Yet, in other disorders, especially retinal vascular disorders, it may be helpful to order additional tests (e.g. fluorescein angiogram).
There are 84,495 X-Ray images (JPEG) and 4 categories (NORMAL,CNV,DME,DRUSEN).
Images are labelled as (disease)-(randomized patient ID)-(image number by this patient) and split into 4 directories:
Optical coherence tomography (OCT) images (Spectralis OCT, Heidelberg Engineering, Germany) were selected from retrospective cohorts of adult patients from the Shiley Eye Institute of the University of California San Diego, the California Retinal Research Foundation, Medical Center Ophthalmology Associates, the Shanghai First People’s Hospital, and Beijing Tongren Eye Center between July 1, 2013 and March 1, 2017.
Before training, each image went through a tiered grading system consisting of multiple layers of trained graders of increasing expertise for verification and correction of image labels. Each image imported into the database started with a label matching the most recent diagnosis of the patient. The first tier of graders consisted of undergraduate and medical students who had taken and passed an OCT interpretation course review. This first tier of graders conducted initial quality control and excluded OCT images containing severe artefacts or significant image resolution reductions. The second tier of graders consisted of four ophthalmologists who independently graded each image that had passed the first tier. The presence or absence of choroidal neovascularization (active or in the form of subretinal fibrosis), macular edema, drusen, and other pathologies visible on the OCT scan were recorded. Finally, the third tier of two senior independent retinal specialists, each with over 20 years of clinical retina experience, verified the true labels for each image. The dataset selection and stratification process are displayed in a CONSORT-style diagram in the above Figure. To account for human error in grading, a validation subset of 993 scans was graded separately by two ophthalmologist graders, with disagreement in clinical labels arbitrated by a senior retinal specialist.
For additional information: Link
With the help of OpenCV’s Imread function, the image is converted to a tensor.
Since the data is extracted sequentially through each class-folder we need to shuffle the data.
we should also separate the labels and the image array.
We need some insights from the data both the classes and the OCT images.
In this step, we are normalizing the data and one-hot encoding the labels so that we can fit them into the neural network.
There were quite a few number of architectures used including some of the pre-trained models.
The results from the custom model were astonishing. Let’s see some of the models used in this project.
In order to check the worst case, it is always good to use a random model before trying out other models.
With the concept of Transfer Learning using pre-trained weights on the image-net, I have built a couple of models in order to increase the performance.
Though there is a myth that the pre-trained models with the help of transfer learning may work better than other models. Here we see that a custom model can perform better than pre-trained models. We can conclude that the type of data which we work on, is far more significant than the models we build.
Retinal Damage Detection by Image-Based Deep Learning
Research & References of Retinal Damage Detection by Image-Based Deep Learning|A&C Accounting And Tax Services
Source
0 Comments