Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LeDNet: Localization-enabled Deep Neural Network for Multi-Label Radiography Image Classification

Lalit Pant, Shubham Arora

TL;DR

LeDNet tackles multi-label thoracic disease classification in chest X-ray images by incorporating lung-region localization to suppress irrelevant background. The method trains a UNet-based lung localization on JSRT, overlays masks onto CheXpert images to produce lung-focused inputs, and uses DenseNet-121 with binary cross-entropy to classify 14 CheXpert observations. Localization improved disease detection compared to using original CheXpert images, with IoU/Dice metrics validating segmentation quality and overlay-based inputs delivering better classification results on CheXpert. The work highlights practical considerations like computation and suggests future directions, including combining features from both source types and applying self-training and ensembles to handle label uncertainty and class imbalance.

Abstract

Multi-label radiography image classification has long been a topic of interest in neural networks research. In this paper, we intend to classify such images using convolution neural networks with novel localization techniques. We will use the chest x-ray images to detect thoracic diseases for this purpose. For accurate diagnosis, it is crucial to train the network with good quality images. But many chest X-ray images have irrelevant external objects like distractions created by faulty scans, electronic devices scanned next to lung region, scans inadvertently capturing bodily air etc. To address these, we propose a combination of localization and deep learning algorithms called LeDNet to predict thoracic diseases with higher accuracy. We identify and extract the lung region masks from chest x-ray images through localization. These masks are superimposed on the original X-ray images to create the mask overlay images. DenseNet-121 classification models are then used for feature selection to retrieve features of the entire chest X-ray images and the localized mask overlay images. These features are then used to predict disease classification. Our experiments involve comparing classification results obtained with original CheXpert images and mask overlay images. The comparison is demonstrated through accuracy and loss curve analyses.

LeDNet: Localization-enabled Deep Neural Network for Multi-Label Radiography Image Classification

TL;DR

LeDNet tackles multi-label thoracic disease classification in chest X-ray images by incorporating lung-region localization to suppress irrelevant background. The method trains a UNet-based lung localization on JSRT, overlays masks onto CheXpert images to produce lung-focused inputs, and uses DenseNet-121 with binary cross-entropy to classify 14 CheXpert observations. Localization improved disease detection compared to using original CheXpert images, with IoU/Dice metrics validating segmentation quality and overlay-based inputs delivering better classification results on CheXpert. The work highlights practical considerations like computation and suggests future directions, including combining features from both source types and applying self-training and ensembles to handle label uncertainty and class imbalance.

Abstract

Multi-label radiography image classification has long been a topic of interest in neural networks research. In this paper, we intend to classify such images using convolution neural networks with novel localization techniques. We will use the chest x-ray images to detect thoracic diseases for this purpose. For accurate diagnosis, it is crucial to train the network with good quality images. But many chest X-ray images have irrelevant external objects like distractions created by faulty scans, electronic devices scanned next to lung region, scans inadvertently capturing bodily air etc. To address these, we propose a combination of localization and deep learning algorithms called LeDNet to predict thoracic diseases with higher accuracy. We identify and extract the lung region masks from chest x-ray images through localization. These masks are superimposed on the original X-ray images to create the mask overlay images. DenseNet-121 classification models are then used for feature selection to retrieve features of the entire chest X-ray images and the localized mask overlay images. These features are then used to predict disease classification. Our experiments involve comparing classification results obtained with original CheXpert images and mask overlay images. The comparison is demonstrated through accuracy and loss curve analyses.
Paper Structure (7 sections, 7 figures)

This paper contains 7 sections, 7 figures.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Methodology
  4. Experimental Results
  5. Experimental Evaluation
  6. Future Work
  7. Conclusion

Figures (7)

  • Figure 1: The Pre-process and Localization Training module
  • Figure 2: The Localization Prediction and Overlay module
  • Figure 3: The pre-processing module to classify empty and uncertain labels
  • Figure 4: The Feature Selection and Prediction module
  • Figure 5: Accuracy Plots and Loss Plots for DenseNet-121 on CheXpert original and overlay images
  • ...and 2 more figures