Computational Biology of Diabetes
Diabetes Mellitus is an abnormal condition resulting from high blood sugar levels which lead to life-threatening diseases like kidney failure, blindness, lower limb amputation, cardiovascular diseases, etc. Diabetic Peripheral Neuropathy is a type of Diabetic Neuropathy and it causes by nerve damage that typically affects the feet and legs and sometimes affects the hands and arms. This type of neuropathy is common and about one third to one-half of people with diabetes have peripheral neuropathy. Diabetic wounds are a major complication of Diabetic Peripheral Neuropathy. Early detection of these wounds will help to save the part of the body and the identification of the severity level will help to define correct treatments.
Particularly, the existing solutions require a well-trained clinician to manually evaluate diabetic clinical images. Though this current solution is effective, it is time-consuming and relies heavily on the expertise of well-training practitioners. Therefore, early detection of peripheral neuropathy will help to protect parts of the body and the identification of the severity level will benefit for precise treatments. There have been no previous studies done on the automated image analysis and grading system. Therefore, it is essential to map clinical images in order to analyse the severity of the wound. Deep learning and image processing techniques are particularly useful in that respect as they have been proven to perform well on related tasks because they learn from data on how to define the severity level of the wound image.
The research addresses a classification model to classify the severity of wounds in diabetes using a deep convolutional neural network (DCNN) and develop a segmentation model to detect the wound boundary and the affected area through an instance-based segmentation technique. The aim is to automate the segmentation and classification of diabetic foot ulcers imagery through deep learning.
In another perspective, the automatic detection of diabetic retinopathy (DR) is of vital importance, as it is the leading cause of irreversible vision loss in the working-age population all over the world today. Basically, DR affects blood vessels in the light-sensitive tissue (i.e. retina). In other words, there are different kinds of abnormal lesions caused by diabetic retinopathy that are important for classifying whether images show signs of retinopathy.
The present solutions require a well-trained clinician to manually evaluate digital colour fundus photographs of retina, and DR is identified by locating the lesions associated with vascular abnormalities due to diabetes. Though this current solution is effective, it is time-consuming and relies heavily on the expertise of ophthalmologists and well-training practitioners. Therefore, early detection of diabetic retinopathy occurrence and similar case(s) retrieval can be very helpful for clinical treatment and decision support process; although several different feature extraction approaches have been proposed, the classification and content-based similar images retrieval tasks for retinal images is still tedious even for those trained clinicians.
Recently, deep convolutional neural networks have manifested superior performance in image classification and content-based image retrieval particularly in the biomedical field compared to previous handcrafted feature-based image classification and retrieval methods.
The research addresses the design and development of a classification model to detect diabetic retinopathy and predict its severity level in retinal images using an ensemble of deep convolutional neural networks (DCNNs). Moreover, this model is extended to a retrieval system to retrieve the k most relevant (Top k) retinal images from a database of retinal images with relevant diagnosis and symptoms for a given query retinal image.