A Novel Deep Learning Technique for Medical Image Analysis Using Improved Optimizer
Vertika Agarwal, Manoj Chandra Lohani, Qurotul Aini, Nesti Anggaraini Santoso · Health Informatics Journal · 2024
Application of Convolutional neural network in spectrum of Medical image analysis are providing benchmark outputs which converges the interest of many researchers to explore it in depth. Availability of clean dataset is one of the significant factor which drives the efficiency of CNN (Convolutional neural network) model. Latest preprocessing technique Real Esrgan and GFPGAN (Generative facial prior GAN) are proving their mettle in providing high resolution dataset. Optimizer also plays a vital role in upgrading the functioning of CNN model. A new optimization technique Gradient Centralization is providing the unparalleled result in terms of generalization and execution time. Our research delves deep into aforementioned factors. Result has already been achieved by taking in account the first factor which is preprocessing technique. In continuation of our effort, our paper explores the next factor which is the employment of new optimization technique, Gradient centralization to our integrated framework (Model with advanced preprocessing technique). Preprocessing integrated Densenet 201, and NasNet are giving encouraging result in terms of training loss and execution time when GC is used as optimizer. Experiments are carried out for three different types of datasets: Skin, Retina and Lung and comparison of loss curves for both preprocessing integrated models is carried out.