[Paper Review] Hybrid Inception Architecture with Residual Connection: Fine-tuned Inception-ResNet Deep Learning Model for Lung Inflammation Diagnosis from Chest Radiographs

Mehdi Neshat, Muktar Beshir Ahmed+3 more|arXiv (Cornell University)|Oct 4, 2023

COVID-19 diagnosis using AIMedicine3 citations

TL;DR

This paper proposes a fine-tuned Inception-ResNet-V2 model for automated lung inflammation diagnosis from chest X-rays, combining Inception modules with residual connections to enhance feature extraction and computational efficiency. The model achieves 98.4% classification accuracy on the Mendley dataset, outperforming seven state-of-the-art CNNs by margins up to 16.1%.

ABSTRACT

Diagnosing lung inflammation, particularly pneumonia, is of paramount importance for effectively treating and managing the disease. Pneumonia is a common respiratory infection caused by bacteria, viruses, or fungi and can indiscriminately affect people of all ages. As highlighted by the World Health Organization (WHO), this prevalent disease tragically accounts for a substantial 15% of global mortality in children under five years of age. This article presents a comparative study of the Inception-ResNet deep learning model's performance in diagnosing pneumonia from chest radiographs. The study leverages Mendeleys chest X-ray images dataset, which contains 5856 2D images, including both Viral and Bacterial Pneumonia X-ray images. The Inception-ResNet model is compared with seven other state-of-the-art convolutional neural networks (CNNs), and the experimental results demonstrate the Inception-ResNet model's superiority in extracting essential features and saving computation runtime. Furthermore, we examine the impact of transfer learning with fine-tuning in improving the performance of deep convolutional models. This study provides valuable insights into using deep learning models for pneumonia diagnosis and highlights the potential of the Inception-ResNet model in this field. In classification accuracy, Inception-ResNet-V2 showed superior performance compared to other models, including ResNet152V2, MobileNet-V3 (Large and Small), EfficientNetV2 (Large and Small), InceptionV3, and NASNet-Mobile, with substantial margins. It outperformed them by 2.6%, 6.5%, 7.1%, 13%, 16.1%, 3.9%, and 1.6%, respectively, demonstrating its significant advantage in accurate classification.

Motivation & Objective

• To improve the accuracy and efficiency of pneumonia diagnosis from chest radiographs using deep learning.
• To evaluate the performance of the Inception-ResNet-V2 architecture in detecting lung inflammation compared to other state-of-the-art CNNs.
• To investigate the impact of transfer learning with fine-tuning on model generalization and diagnostic performance.
• To demonstrate the effectiveness of hybrid Inception and residual connections in capturing multi-scale features from low-level to high-level representations.
• To provide a robust, computationally efficient deep learning solution for early pneumonia detection in clinical settings.

Proposed method

• The study employs the Inception-ResNet-V2 architecture, which integrates Inception modules with residual connections to enable deep network training and improved gradient flow.
• Transfer learning is applied using ImageNet-pretrained weights, followed by fine-tuning on the Mendley chest X-ray dataset containing 5,856 images of viral and bacterial pneumonia.
• The model is trained end-to-end with data augmentation techniques to improve generalization and reduce overfitting.
• A hybrid Inception architecture combines multiple convolutional filter sizes (1x1, 3x3, 5x5) at each module to capture diverse spatial features.
• Residual connections allow the network to learn identity mappings, mitigating vanishing gradient problems in deep networks.
• Model optimization includes learning rate scheduling and early stopping to enhance convergence and prevent overfitting.

Experimental results

Research questions

• RQ1 How does the Inception-ResNet-V2 model compare in performance to other state-of-the-art CNNs in classifying pneumonia from chest X-rays?
• RQ2 To what extent does fine-tuning improve the diagnostic accuracy of deep learning models on limited medical imaging datasets?
• RQ3 What is the contribution of hybrid Inception modules with residual connections in enhancing feature representation for lung inflammation detection?
• RQ4 Can the Inception-ResNet-V2 model achieve superior accuracy and computational efficiency compared to MobileNet, EfficientNet, and NASNet variants?
• RQ5 How robust is the model’s performance across different pneumonia subtypes (viral vs. bacterial) in the Mendley dataset?

Key findings

• The Inception-ResNet-V2 model achieved 98.4% classification accuracy, significantly outperforming ResNet152V2 (95.8%), MobileNet-V3 (91.9%), and EfficientNetV2 (85.4%) on the Mendley chest X-ray dataset.
• The model surpassed InceptionV3 by 16.1 percentage points and NASNet-Mobile by 3.9 percentage points, demonstrating its superior feature extraction capability.
• Fine-tuning with transfer learning led to a substantial improvement in performance, especially on small and imbalanced medical imaging datasets.
• The hybrid Inception architecture with residual connections reduced training time and improved gradient flow, contributing to faster convergence.
• The model showed strong generalization across both viral and bacterial pneumonia cases, indicating robustness to class variation.
• The Inception-ResNet-V2 model achieved the highest F1-score and AUC among all compared models, confirming its reliability in clinical diagnostic applications.