Scribd
Upload
34 views11 pages

Detection of Tuberculosis Based On Deep Learning Methods PPT

Detection of tuberculosis

Uploaded by

www.ayethetzaw
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
34 views11 pages

Detection of Tuberculosis Based On Deep Learning Methods PPT

Detection of tuberculosis

Uploaded by

www.ayethetzaw
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 11

Detection of Tuberculosis based

on Deep Learning Methods


by Murali Krishna Puttagunta and S. Ravi 2021 J. Phys.: Conf. Ser. 1767 012004

Presented by:
6CT-2 Mg Aye Thet Zaw
Abstract
• Pulmonary Tuberculosis (TB): A serious, easily transmissible disease.
• Deep Learning in Diagnostics: Deep learning, a successful technique for analyzing medical images, is used to
help diagnose TB from chest X-rays.
• Convolutional Neural Networks (CNNs): A common deep learning method used for TB detection.
Introduction
• Computer-aided diagnostic (CAD) systems : have been in development since the 1970s.
• Advancements in AI: CAD systems classify abnormalities in radiological images.
• Applications of CAD Systems: Alzheimer's disease, breast cancer, prostate cancer, lung cancer, neurological
disorders, bone issues, and skin lesions.
• AI Methods Used: Machine Learning (ML) and Deep Learning (DL) for analyzing chest X-rays (CXRs).

Figure 1. Computer-aided diagnostic (CAD) system


Deep Learning
• Convolutional Neural Networks (CNNs): CNNs are crucial for detecting and classifying TB in chest X-rays.
•Convolutional Layers: Extract features from images using convolution operations.
•Pooling Layers: Reduce the dimensions of feature maps (e.g., max pooling and average pooling) and help
prevent overfitting.
•Fully Connected (FC) Layers: Use extracted features to classify images into different categories.
Deep Learning Cont.
• Activation Functions: Functions like softmax are used in the FC layers to generate the final output.
• ImageNet Dataset: A large dataset with over 15 million images and nearly 22,000 categories, commonly
used for testing image classification models. The ImageNet Large Scale Visual Recognition Challenge (ILSVRC)
uses a subset of 1,000 categories.
Deep Learning Cont.
CNN Variants: Several CNN models have been developed, including:
•LeNet: Early model with convolutional, pooling, and FC layers.
Deep Learning Cont.
•AlexNet: Introduced ReLU activation function, consisting of 5 convolutional and 3 FC layers.

Figure 5. A Basic CNN Architecture using AlexNet


Deep Learning Cont.
•GoogleNet: Uses inception layers with different filter sizes.

Figure 5. A Basic CNN Architecture using A GoogleNet


Deep Learning Cont.
•ResNet: Features deep networks with shortcut connections to address vanishing gradient problems.

Figure 5. A Basic CNN Architecture using ResNet

Training CNNs: Requires large amounts of data to prevent overfitting. In the medical field, where data may be
limited, transfer learning is used:
•Feature Extraction: Using a pre-trained CNN model to extract features.
•Fine-Tuning: Adjusting a pre-trained model with data from a specific domain.
Dataset
1.TBX11K Dataset:
•Contains 11,200 X-ray images.
•Includes 1,200 images with TB, 5,000 sick but non-TB, and 5,000 healthy images.
•Resolution: 3000 × 3000 pixels.
2.JSRT Dataset:
•Developed by the Japanese Radiological Technology Society (JSRT) and the Japanese Radiological Society (JRS).
•Includes 154 X-ray images with lung nodules and 93 normal images.
•Resolution: 2048 × 2048 pixels.
3.Montgomery Dataset (MC):
•Collected from a tuberculosis prevention program in Montgomery County, USA.
•Contains 80 normal X-ray images and 58 with TB manifestations.
•Resolution: 4020 × 4892 or 4892 × 4020 pixels.
4.Shenzhen Dataset (CH):
•Obtained from Shenzhen No.3 Hospital, China.
•Includes 326 normal X-ray images and 336 with TB.
•Resolution: 3000 × 3000 pixels.
5.India Dataset (IN):
•Comprises two subsets, DA and DB, from the National Institute for Tuberculosis and Respiratory Diseases, New Delhi.
•DA: Training set with 52 CXRs with and without TB, and test set with 26 CXRs each.
•DB: Training set with 50 CXRs with and without TB, and test set with 25 CXRs each.
6.Belarus Dataset:
•Includes 420 X-ray images and chest CT scans of TB patients.
•Resolution: 2248 × 2248 pixels.
•Compiled by various institutes under the Ministry of Health of the Republic of Belarus.
Conclusion
This paper gives a quick overview for new researchers interested in using deep learning to detect TB
in chest X-rays. CAD systems help spot TB by analyzing these X-rays. However, none of these systems have
been approved by the World Health Organization (WHO) because more proof is needed to confirm their
effectiveness. This review shows that more research is needed on CAD systems for TB detection. There is a lot
of potential for future research in this area, and AI-based CAD systems will likely become very important in
detecting TB.

You might also like