What is it about?
The research talks about how important it is to correctly identify certain features in X-ray images for medical purposes. Doctors use different techniques like X-rays, CT-Scans, and ultrasound to look at the insides of the human body without needing surgery. But training a computer to do this task from scratch is really hard because it takes a lot of time, a lot of labeled data, and a lot of experience. So, instead, it's better to start with a computer model that's already been trained on a lot of medical data and then fine-tune it for the specific task. This study compared two such pre-trained models and also tried to prevent the computer from memorizing the training data instead of actually learning from it. The study found that with the right fine-tuning, these pre-trained models performed just as well as a model that was trained from scratch.
Featured Image
Photo by Markus Spiske on Unsplash
Why is it important?
This is important for several reasons: Efficiency: Training a deep learning model from scratch requires substantial computational resources and time. Using a pre-trained model accelerates this process significantly because it has already learned useful features from the vast amount of data it was initially trained on. Data scarcity: In many specialized fields like medical imaging, obtaining a large labeled dataset for training is often challenging due to privacy concerns, rarity of certain conditions, and the need for expert annotation. Pre-trained models, having been trained on extensive datasets, can help overcome this limitation. Performance: Pre-trained models often perform as well, if not better, than models trained from scratch, especially when the available data for a specific task is limited. So, the ability to use and fine-tune pre-trained models allows for the application of advanced machine learning techniques in specialized fields, even when resources or data are limited.
Perspectives
This publication makes a significant contribution to the field of medical imaging analysis. It acknowledges the challenges in training deep learning models from scratch - the demand for computational resources, extensive labeled data, and the need for domain experience. The finding that fine-tuned pre-trained models perform comparably to those trained from scratch is encouraging. It suggests that leveraging pre-trained models could be a practical and efficient approach in medical imaging scenarios, which often grapple with limited data and resources.
Mr Victor Ikechukwu Agughasi
Maharaja Institute of Technology
Read the Original
This page is a summary of: ResNet-50 vs VGG-19 vs training from scratch: A comparative analysis of the segmentation and classification of Pneumonia from chest X-ray images, Global Transitions Proceedings, November 2021, Elsevier,
DOI: 10.1016/j.gltp.2021.08.027.
You can read the full text:
Resources
Contributors
The following have contributed to this page
