Loading [a11y]/accessibility-menu.js
Transfer Learning with Fine-Tuned MobileNetV2 for Diabetic Retinopathy | IEEE Conference Publication | IEEE Xplore

Transfer Learning with Fine-Tuned MobileNetV2 for Diabetic Retinopathy


Abstract:

Blindness is a vision impairment that cannot be corrected fully with medication or surgery or through glasses. One of the reason for blindness is Diabetic Retinopathy. It...Show More

Abstract:

Blindness is a vision impairment that cannot be corrected fully with medication or surgery or through glasses. One of the reason for blindness is Diabetic Retinopathy. It is a medical condition that damages eye retinal tissues. In todays era emphasis is on finding automatic computational mechanism that can check the severity of diabetic retinopathy so that the blindness can be detected before it happened. Instead of designing a deep neural network from scratch this paper proposes an approach based on transfer learning. MobileNetv2, a predefined model is used for extracting a meaningful features from the given set of retina images. Model is customized by adding the globalaveragepooling layer and softmax classifier layer on the top of pretrained base model for classifying images in one of the five different classes of diabetic retinopathy. Initially we have only trained the stacked layers, thereby refraining the weights of pretrained model from updation. To further improve the performance of the model the weights of top few layers of the base network are fine-tuned. The network now adapts to itself with these specialized features. Kaggle diabetic retinopathy dataset is used for evaluating the performance of the proposed approach. 2929 retinal fundus training images, 733 validation images are used to build the model and tested with 1928 testing images. Experimental result shows that after fine tuning the network training accuracy increased from 70% to 91% and validation accuracy increased from 50% to 81%. Training loss and validation loss is observed to be approximately same, that indicates model is perfectly fit. This accuracy of fine-tuned network reveals a noticeable improvement.
Date of Conference: 05-07 June 2020
Date Added to IEEE Xplore: 03 August 2020
ISBN Information:
Conference Location: Belgaum, India

I. Introduction

In todays digital era, a massive amount of data is continuously being generated at remarkable and ever increasing scales. Studies shows that tremendous amount of data is being collected in various domain such as in biomedical engineering research, social networks, security, etc.[1]. It is difficult for human being to process this significant amount of data in a particular domain within a stipulated time. Machine learning, a sub domain of artificial intelligence have been applied in almost every domain to provide possible solutions. Deep learning, a sub field of machine learning techniques uses artificial neural network architecture. It uses datasets to extract features and has the automatic learning ability [2]. Deep learning, is getting lots of attention as it is possible to achieve human-level performance which were not possible before [2]. It is a multilayer architecture where in the network can have as many as 150 hidden layers. The input layer is used to provide input data, each hidden layer progressively extract features from the data and output layer is used to give desired result. Deep Convolutional Neural Network is well suited for analyzing and classification of medical images. Study reveals that building a small and efficient network is a need of an hour. With the increasing usage of mobile devices in day to day life and the success stories of deep learning techniques in various domain motivates the researchers to use deep learning algorithms on mobile devices for providing users with intelligent services [3]. MobileNets, a small network matches the restriction of limited resources but optimizes the latency. MobileNets are especially designed for mobile and vision application [4]. In medical imaging field, it is difficult to have sufficient amount of image dataset to train a Deep Convolution Neural Network from scratch. Training a network from scratch requires not only extensive computational and memory resources but also large datasets. Training a network with small dataset quite often leads to overfitting problem [5]. Transfer learning is a solution for this. In this paper transfer learning approach is used where in MobileNetv2, a pretrained model is fine tuned for Diabetic Retinopathy classification. Diabetic Retinopathy is an eye disease resulting from diabetes and have impact on health system. [6]. If not treated at early stage it may lead to low vision and blindness. Diagnosis of DR requires frequent clinical visit and ophthalmologist opinion. The accuracy and mobility of medical examination equipment is also important. MobileNetv2 model is an efficient network for mobile vision application can be finetuned for training task on fundus datasets and later can be used as an inference engine as an when required to infer the required information when provided with new fundus image.

Contact IEEE to Subscribe

References

References is not available for this document.