Application of Deep Learning Techniques for Automated Healthcare Data Labelling

Subject Area: Medical Image Labelling

This paper on automated data labeling for health care using integrated deep learning technique was aimed at developing a software application system in medical image classification. To achieve this, data of health care images was collected from Nigerian hospitals considering four classes of eye conditions which are cataract, glaucoma, diabetes and normal eye. The data was extracted with the application of Mobile-Net version 2, which is a deep learning algorithm carefully designed with series of convolutional filters with the ability to extract spatial and intricate image information for training purposes. The extracted feature vectors were formed as a data matrix using global average pooling layer and then feed to a clustered based neural network algorithm and trained with back-propagation to generate a model for health care classification and labeling. During the training of the neurons, Self-Organized Map (SOM) were formulated through the adjustment of the neurons and clustering similar data points in a grid map, until convergence. The model generated was deployed as a software application for diagnosis of eye diseases using Tensorflow and MATLAB programming language and then evaluated using real-life data of health care images.