INTELLIGENT MEDICAL IMAGE DATA LABELLING HEALTHCARE AUTOMATION USING DEEP LEARNING TECHNIQUE

Subject Area: Medical Image Data Labelling

The goal of this paper was to create a software application system medical image classification through automated data labelling for healthcare utilising an integrated deep learning technique. The four categories of eye conditions such as cataract, glaucoma, diabetes, and normal eyewere taken into consideration when gathering health care imaging data from Nigerian hospitals. The deployment of Mobile-Net-2, a deep learning technique with a number of convolutional filters that can extract complex and spatial visual information for training, was used to extract the data.In order to create a model for health care classification and labelling, the collected feature vectors were first converted into a data matrix using a global average pooling layer. They were then fed into a clustering neural network algorithm and trained using back-propagation. Self-Organised Maps (SOMs), were created throughout the neurone training process by adjusting the neurones and grouping related data points together in a grid map until they converged. Using Tensorflow and the MATLAB programming language, the model was implemented as a software program for diagnosing eye conditions. It was then assessed using actual medical imaging data. When accuracy was taken into account, the findings showed that the condition was 97.039% for normal eye data, 99.421% for diabetic retinopathy, 90.926% for glaucoma, and 98.431% for cataracts.To validate the results, comparative analysis with other algorithms was performed and the results showed that the new model with Mobile-Net and Clustered based neural network achieved the best results.