Three-Dimensional Semantic Segmentation of Diabetic Retinopathy Lesions and Grading Using Transfer Learning DOI Open Access
Natasha Shaukat, Javeria Amin, Muhammad Sharif

et al.

Journal of Personalized Medicine, Journal Year: 2022, Volume and Issue: 12(9), P. 1454 - 1454

Published: Sept. 5, 2022

Diabetic retinopathy (DR) is a drastic disease. DR embarks on vision impairment when it left undetected. In this article, learning-based techniques are presented for the segmentation and classification of lesions. The pre-trained Xception model utilized deep feature extraction in phase. extracted features fed to Deeplabv3 semantic segmentation. For training model, an experiment performed selection optimal hyperparameters that provided effective results testing multi-classification developed using fully connected (FC) MatMul layer efficient-net-b0 pool-10 squeeze-net. from both models fused serially, having dimension N × 2020, amidst best 1032 chosen by applying marine predictor algorithm (MPA). lesions into grades 0, 1, 2, 3 neural network KNN classifiers. proposed method performance validated open access datasets such as DIARETDB1, e-ophtha-EX, IDRiD, Messidor. obtained better compared those latest published works.

Language: Английский

An Empirical Evaluation of Enhanced Performance Softmax Function in Deep Learning DOI Creative Commons

Sumiran Mehra,

Gopal Raut,

R. D. Purkayastha

et al.

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 34912 - 34924

Published: Jan. 1, 2023

This article presents an enhanced-performance, hardware-efficient Softmax Function (SF) for a deep neural network accelerator. is used in the classification layer learning models and also hidden layers of advanced networks like Transformer Capsule networks. The major challenge designing efficient hardware architecture SF complex exponential division computational sub-blocks. Utilizing mutual exclusivity CO-ordinate Rotational DIgital Computer (CORDIC) algorithm, hardware-optimized pipelined CORDIC-based considered area, power, enhanced throughput design. In order to maintain good accuracy models, proposed design undergoes Pareto study on variation number pipeline stages. quantized 16-bit precision, inference validated various datasets. prototyped using Xilinx Zynq FPGA can be operated at 685MHz. Also, ASIC implementation performed 45nm technology node 5GHz maximum operating frequency. achieves validation loss less than 2% account reduced silicon area Energy-Delay-Product(EDP) (by 12×). Post synthesis simulation result illustrates that 3× better performance terms logic delay compared state-of-the-art architectures.

Language: Английский

Citations

16
Revolutionizing Dynamic Electric Vehicle Charging: Innovations in Inductive Power Transfer System Optimization DOI
S Vikram Singh, Sanjeev Kumar Sharma, Amit Dutt

et al.

Published: May 9, 2024

Language: Английский

Citations

6
GNN-fused CapsNet with multi-head prediction for diabetic retinopathy grading DOI

Yongjia Lei,

Shuyuan Lin, Zhiying Li

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2024, Volume and Issue: 133, P. 107994 - 107994

Published: Feb. 7, 2024

Language: Английский

Citations

5
A Deep Learning Model for Detecting Diabetic Retinopathy Stages with Discrete Wavelet Transform DOI Creative Commons

A. M. Mutawa,

Khalid Al-Sabti,

Seemant Raizada

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(11), P. 4428 - 4428

Published: May 23, 2024

Diabetic retinopathy (DR) is the primary factor leading to vision impairment and blindness in diabetics. Uncontrolled diabetes can damage retinal blood vessels. Initial detection prompt medical intervention are vital preventing progressive impairment. Today’s growing field presents a more significant workload diagnostic demands on professionals. In proposed study, convolutional neural network (CNN) employed detect stages of DR. This research crucial for studying DR because its innovative methodology incorporating two different public datasets. strategy enhances model’s capacity generalize unseen images, as each dataset encompasses unique demographics clinical circumstances. The learn capture complicated hierarchical image features with asymmetric weights. Each preprocessed using contrast-limited adaptive histogram equalization discrete wavelet transform. model trained validated combined datasets Dataset Retinopathy Asia-Pacific Tele-Ophthalmology Society. CNN tuned learning rates optimizers. An accuracy 72% an area under curve score 0.90 was achieved by Adam optimizer. recommended study results may reduce diabetes-related early identification severity.

Language: Английский

Citations

5
Three-Dimensional Semantic Segmentation of Diabetic Retinopathy Lesions and Grading Using Transfer Learning DOI Open Access
Natasha Shaukat, Javeria Amin, Muhammad Sharif

et al.

Journal of Personalized Medicine, Journal Year: 2022, Volume and Issue: 12(9), P. 1454 - 1454

Published: Sept. 5, 2022

Diabetic retinopathy (DR) is a drastic disease. DR embarks on vision impairment when it left undetected. In this article, learning-based techniques are presented for the segmentation and classification of lesions. The pre-trained Xception model utilized deep feature extraction in phase. extracted features fed to Deeplabv3 semantic segmentation. For training model, an experiment performed selection optimal hyperparameters that provided effective results testing multi-classification developed using fully connected (FC) MatMul layer efficient-net-b0 pool-10 squeeze-net. from both models fused serially, having dimension N × 2020, amidst best 1032 chosen by applying marine predictor algorithm (MPA). lesions into grades 0, 1, 2, 3 neural network KNN classifiers. proposed method performance validated open access datasets such as DIARETDB1, e-ophtha-EX, IDRiD, Messidor. obtained better compared those latest published works.

Language: Английский

Citations

21