Related Experiment Video
Updated: Jun 26, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
A novel deep learning technique for medical image analysis using improved optimizer
Vertika Agarwal1, M C Lohani1, Ankur Singh Bist1
1Graphic Era Hill University, India.
Health Informatics Journal
|May 17, 2024
Summary
This study introduces Gradient Centralization (GC) as a novel optimizer for medical image analysis. Integrating GC with advanced preprocessing techniques like Real ESRGAN and GFPGAN significantly improves deep learning model performance, reducing execution time and loss factors.
Area of Science:
- Medical Image Analysis
- Deep Learning
- Computer Vision
Background:
- Convolutional Neural Networks (CNNs) are crucial for medical image analysis, but their performance is often limited by optimizer efficiency and memory requirements.
- Existing optimizers like Stochastic Gradient Descent face challenges such as high variance and computational expense.
- Advanced preprocessing techniques like Real ESRGAN and GFPGAN enhance image resolution, but their integration with efficient optimizers is key.
Purpose of the Study:
- To explore the efficacy of Gradient Centralization (GC) as a novel optimization technique for CNNs in medical image analysis.
- To evaluate the performance of an integrated framework combining advanced preprocessing (Real ESRGAN, GFPGAN) with GC.
- To address limitations of traditional optimizers, including slow processing, high memory usage, and the dead neuron problem.
Main Methods:
- Implementation of an integrated framework utilizing Real ESRGAN and GFPGAN for high-resolution medical image dataset generation.
- Integration of the Gradient Centralization (GC) optimization technique within the deep learning model.
- Comparative analysis of the proposed framework against traditional optimization methods in terms of execution time and loss reduction.
Main Results:
- The integrated framework demonstrated significant improvements in execution time compared to conventional optimizers.
- Gradient Centralization effectively reduced the loss factor, indicating enhanced model generalization.
- The combination of advanced preprocessing and GC provided an optimal solution for deep learning in medical imaging.
Conclusions:
- Gradient Centralization offers a promising optimization strategy for deep learning models in medical image analysis.
- The integrated framework of Real ESRGAN, GFPGAN, and GC presents a robust solution for improving model efficiency and accuracy.
- Further research into GC and its application with various deep learning architectures is warranted.
