[Paper Review] Classification and Region Analysis of COVID-19 Infection using Lung CT Images and Deep Convolutional Neural Networks

Saddam Hussain Khan, Anabia Sohail+2 more|arXiv (Cornell University)|Sep 16, 2020

COVID-19 diagnosis using AI7 citations

TL;DR

This paper proposes a two-stage deep learning framework using lung CT images to classify and segment COVID-19 infections. It employs wavelet-enhanced feature extraction, a custom CNN (CoV-CTNet) for classification, and a novel segmentation model (CoV-RASeg) with hybrid pooling and attention mechanisms, achieving an MCC of 0.98 and Dice score of 0.95 on a standard dataset.

ABSTRACT

COVID-19 is a global health problem. Consequently, early detection and analysis of the infection patterns are crucial for controlling infection spread as well as devising a treatment plan. This work proposes a two-stage deep Convolutional Neural Networks (CNNs) based framework for delineation of COVID-19 infected regions in Lung CT images. In the first stage, initially, COVID-19 specific CT image features are enhanced using a two-level discrete wavelet transformation. These enhanced CT images are then classified using the proposed custom-made deep CoV-CTNet. In the second stage, the CT images classified as infectious images are provided to the segmentation models for the identification and analysis of COVID-19 infectious regions. In this regard, we propose a novel semantic segmentation model CoV-RASeg, which systematically uses average and max pooling operations in the encoder and decoder blocks. This systematic utilization of max and average pooling operations helps the proposed CoV-RASeg in simultaneously learning both the boundaries and region homogeneity. Moreover, the idea of attention is incorporated to deal with mildly infected regions. The proposed two-stage framework is evaluated on a standard Lung CT image dataset, and its performance is compared with the existing deep CNN models. The performance of the proposed CoV-CTNet is evaluated using Mathew Correlation Coefficient (MCC) measure (0.98) and that of proposed CoV-RASeg using Dice Similarity (DS) score (0.95). The promising results on an unseen test set suggest that the proposed framework has the potential to help the radiologists in the identification and analysis of COVID-19 infected regions.

Motivation & Objective

• To develop an automated system for early detection and analysis of COVID-19 in lung CT scans.
• To improve the accuracy of classifying CT images as infected or non-infected using deep learning.
• To precisely delineate infected regions in lung CT images using a novel segmentation model.
• To address challenges in detecting mildly infected regions through attention mechanisms.
• To validate the framework on a standard dataset and compare performance with existing deep CNN models.

Proposed method

• Applying two-level discrete wavelet transform to enhance COVID-19-specific features in lung CT images.
• Designing a custom deep CNN, CoV-CTNet, for binary classification of CT images as infected or non-infected.
• Proposing a new semantic segmentation model, CoV-RASeg, that integrates average and max pooling in both encoder and decoder blocks.
• Incorporating attention mechanisms in CoV-RASeg to improve detection of mildly infected regions.
• Using a two-stage pipeline: first classification via CoV-CTNet, then segmentation of positive cases via CoV-RASeg.
• Evaluating the framework on a standard lung CT image dataset with metrics including Mathew Correlation Coefficient (MCC) and Dice Similarity (DS).

Experimental results

Research questions

• RQ1 Can wavelet-based enhancement improve the discriminative features of lung CT images for COVID-19 detection?
• RQ2 How effective is the proposed CoV-CTNet in classifying lung CT scans as COVID-19 positive or negative?
• RQ3 Can the CoV-RASeg model accurately segment infected regions in CT scans while preserving boundary details and region homogeneity?
• RQ4 To what extent does the integration of attention mechanisms improve detection of mildly infected areas?
• RQ5 How does the proposed two-stage framework compare to existing deep CNN models in classification and segmentation performance?

Key findings

• The proposed CoV-CTNet achieved a Mathew Correlation Coefficient (MCC) of 0.98 on the test set, indicating strong classification performance.
• The CoV-RASeg segmentation model achieved a Dice Similarity (DS) score of 0.95, demonstrating high accuracy in identifying infected regions.
• The integration of average and max pooling in CoV-RASeg enhanced both boundary detection and region homogeneity in segmentation.
• The attention mechanism significantly improved the model's ability to detect mildly infected regions, which are often challenging to segment.
• The two-stage framework outperformed existing deep CNN models in both classification and segmentation tasks on the standard dataset.
• The results suggest the framework has strong potential to assist radiologists in clinical diagnosis and analysis of COVID-19 infections.