[Paper Review] Advancing Diagnostic Precision: Leveraging Machine Learning Techniques for Accurate Detection of Covid-19, Pneumonia, and Tuberculosis in Chest X-Ray Images

Aditya Kulkarni, Guruprasad Parasnis+4 more|arXiv (Cornell University)|Oct 9, 2023

COVID-19 diagnosis using AIMedicine3 citations

TL;DR

This paper proposes a novel convolutional neural network (CNN) for multiclass detection of COVID-19, pneumonia, and tuberculosis in chest X-ray images using advanced image preprocessing and deep learning. It achieves state-of-the-art performance with AUC scores of 0.95 (COVID-19), 0.99 (tuberculosis), and 0.98 (pneumonia), outperforming transfer learning models like VGG19, ResNet, and DenseNet on multiple public datasets.

ABSTRACT

Lung diseases such as COVID-19, tuberculosis (TB), and pneumonia continue to be serious global health concerns that affect millions of people worldwide. In medical practice, chest X-ray examinations have emerged as the norm for diagnosing diseases, particularly chest infections such as COVID-19. Paramedics and scientists are working intensively to create a reliable and precise approach for early-stage COVID-19 diagnosis in order to save lives. But with a variety of symptoms, medical diagnosis of these disorders poses special difficulties. It is essential to address their identification and timely diagnosis in order to successfully treat and prevent these illnesses. In this research, a multiclass classification approach using state-of-the-art methods for deep learning and image processing is proposed. This method takes into account the robustness and efficiency of the system in order to increase diagnostic precision of chest diseases. A comparison between a brand-new convolution neural network (CNN) and several transfer learning pre-trained models including VGG19, ResNet, DenseNet, EfficientNet, and InceptionNet is recommended. Publicly available and widely used research datasets like Shenzen, Montogomery, the multiclass Kaggle dataset and the NIH dataset were used to rigorously test the model. Recall, precision, F1-score, and Area Under Curve (AUC) score are used to evaluate and compare the performance of the proposed model. An AUC value of 0.95 for COVID-19, 0.99 for TB, and 0.98 for pneumonia is obtained using the proposed network. Recall and precision ratings of 0.95, 0.98, and 0.97, respectively, likewise met high standards.

Motivation & Objective

• To develop a robust, multiclass deep learning model for accurate detection of COVID-19, pneumonia, and tuberculosis from chest X-ray images.
• To improve diagnostic precision by integrating advanced image preprocessing techniques such as histogram equalization and noise removal.
• To outperform existing transfer learning models (e.g., VGG19, ResNet, DenseNet) in classifying chest diseases using public datasets.
• To establish a benchmark for automated chest X-ray analysis in clinical diagnostics through rigorous evaluation across multiple datasets.
• To reduce false positives and negatives in disease detection, enhancing reliability for real-world medical applications.

Proposed method

• A custom convolutional neural network (CNN) was designed with residual connections and global average pooling to enhance feature learning and generalization.
• Image preprocessing included noise removal, histogram equalization, and image scaling to improve input quality and model robustness.
• The model was trained and evaluated on four public datasets: Shenzen, Montgomery, NIH, and a multiclass Kaggle dataset containing 7,395 X-ray images.
• Transfer learning models (VGG19, ResNet, DenseNet, EfficientNet, InceptionNet) were compared against the proposed CNN using identical evaluation protocols.
• Performance was assessed using standard metrics: recall, precision, F1-score, and Area Under Curve (AUC), with histogram equalization identified as the optimal preprocessing step.
• The final model used a softmax classifier for multiclass output and was fine-tuned using data augmentation to improve generalization.

Figure 1: Data Pre-processing by Augmentation

Experimental results

Research questions

• RQ1 Can a custom-designed CNN outperform established transfer learning models in multiclass detection of COVID-19, pneumonia, and tuberculosis from chest X-rays?
• RQ2 How does histogram equalization preprocessing compare to other image enhancement techniques in improving model performance across diverse datasets?
• RQ3 What is the diagnostic accuracy of the proposed model on benchmark datasets such as NIH and multiclass Kaggle?
• RQ4 To what extent does the model reduce false positives and negatives compared to traditional manual diagnosis and prior deep learning approaches?
• RQ5 Can the proposed model generalize across different imaging conditions and patient populations in real-world clinical settings?

Key findings

• The proposed CNN achieved an AUC of 0.95 for COVID-19, 0.99 for tuberculosis, and 0.98 for pneumonia, surpassing all evaluated transfer learning models.
• The model recorded a recall of 0.95, precision of 0.98, and F1-score of 0.97 for COVID-19, indicating high sensitivity and specificity.
• Histogram equalization was identified as the most effective preprocessing technique, significantly enhancing model performance across all datasets.
• On the multiclass Kaggle dataset, the model achieved an AUC of 0.98, demonstrating strong generalization across disease classes.
• The model outperformed ResNet-50 (AUC 0.77) and DenseNet-121 (AUC 0.76) from prior studies, confirming its state-of-the-art performance.
• The model achieved a 99.3% accuracy, 0.99 sensitivity, and 0.99 specificity on a multiclass test set of 7,395 X-ray images, indicating exceptional diagnostic reliability.

Figure 2: Data Pre-processing by LTP