Classifying Chest X-Ray COVID-19 images via Transfer Learning
2021 Ethics and Explainability for Responsible Data Science Conference, EE-RDS 2021
; 2021.
Article
in English
| Scopus | ID: covidwho-1741174
ABSTRACT
The internal behavior of Deep Neural Network architectures can be difficult to interpret. Certain architectures achieve impressive feats in a particular dataset while failing to show comparable performance in other datasets. Developing an architecture that performs well on a dataset can be a time-consuming affair and computationally intensive process. This study explains the effect of transfer learning by fine-tuning already available state-of-the-art architectures in different datasets and using them to classify Chest X-Ray images with high accuracy. Using transfer learning helps the model learn problem-specific features in a short period. It further shows that different models perform differently in a particular setting for a dataset. Ablation studies show that a combination of smaller structures that gives an overall better result may not give the best result in the combined model. In addition, the 'belief' of the model for selecting a particular class is visualized in this study. © 2021 IEEE.
Full text:
Available
Collection:
Databases of international organizations
Database:
Scopus
Language:
English
Journal:
2021 Ethics and Explainability for Responsible Data Science Conference, EE-RDS 2021
Year:
2021
Document Type:
Article
Similar
MEDLINE
...
LILACS
LIS