Effective pneumonia detection using ResNet based transfer learning
Pneumonia is a deadly lungs disease known as silent killer is due to bacterial, viral, or fungal infection and causes lung alveoli to fill with pus or fluids. The most common diagnostic tool for pneumonia is Chest X-rays. However, due to several other medical conditions in the lungs, such as volume...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
The Mattingley Publishing Co., Inc.
2020
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/79986/1/Effective%20Pneumonia%20Detection%20using%20ResNet%20based%20Transfer%20Learning.pdf http://irep.iium.edu.my/79986/ http://www.testmagzine.biz/index.php/testmagzine/article/view/3228/2851 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Pneumonia is a deadly lungs disease known as silent killer is due to bacterial, viral, or fungal infection and causes lung alveoli to fill with pus or fluids. The most common diagnostic tool for pneumonia is Chest X-rays. However, due to several other medical conditions in the lungs, such as volume loss, bleeding,lung cancer,fluid overload,post-radiation or surgery, the diagnosis of pneumonia using chest X-rays becomes very complicated. Therefore, there is a dire need for computer-aided diagnosis systems to assist clinicians in making better decisions. This work proposes an effective, deep convolutional neural network with ResNet-50 architecture for pneumonia detection ResNet has performed quite well on the image recognition task and was a winner of the ImageNet challenge.A pre-trained ResNet-50 model is re-trained with the use of Transfer Learning on two different datasets of chest x-ray images. ResNet-50 based diagnostics model is found useful for pneumonia diagnostics despite significant variations in two datasets. The trained model has achieved an accuracy of 96.76%, which is at par with state-of-the-art techniques available. RSNA dataset, with five times more images than the Chest X-ray Image dataset, took very little time for training. Also, because of the use of the Transfer Learning technique, both the models were able to learn the significant features of pneumonia with only 50% training dataset size.However, the model can be improvised by using more deeper networks. Work can be extended to detect and classify both lung cancer and pneumonia using X-ray images. |
|---|