[Paper Review] Study of Different Deep Learning Approach with Explainable AI for Screening Patients with COVID-19 Symptoms: Using CT Scan and Chest X-ray Image Dataset

Md Manjurul Ahsan, Kishor Datta Gupta+4 more|arXiv (Cornell University)|Jul 24, 2020

COVID-19 diagnosis using AIMedicine66 references38 citations

TL;DR

This study compares eight pre-trained CNNs (VGG16, InceptionResNetV2, ResNet50, DenseNet201, VGG19, MobileNetV2, NasNetMobile, ResNet15V2) on paired CT and chest X-ray datasets for COVID-19 screening, uses LIME for explainability, and identifies NasNetMobile (CT) and NasNetMobile (X-ray) as top performers.

ABSTRACT

The outbreak of COVID-19 disease caused more than 100,000 deaths so far in the USA alone. It is necessary to conduct an initial screening of patients with the symptoms of COVID-19 disease to control the spread of the disease. However, it is becoming laborious to conduct the tests with the available testing kits due to the growing number of patients. Some studies proposed CT scan or chest X-ray images as an alternative solution. Therefore, it is essential to use every available resource, instead of either a CT scan or chest X-ray to conduct a large number of tests simultaneously. As a result, this study aims to develop a deep learning-based model that can detect COVID-19 patients with better accuracy both on CT scan and chest X-ray image dataset. In this work, eight different deep learning approaches such as VGG16, InceptionResNetV2, ResNet50, DenseNet201, VGG19, MobilenetV2, NasNetMobile, and ResNet15V2 have been tested on two dataset-one dataset includes 400 CT scan images, and another dataset includes 400 chest X-ray images studied. Besides, Local Interpretable Model-agnostic Explanations (LIME) is used to explain the model's interpretability. Using LIME, test results demonstrate that it is conceivable to interpret top features that should have worked to build a trust AI framework to distinguish between patients with COVID-19 symptoms with other patients.

Motivation & Objective

• Motivate rapid, scalable screening of COVID-19 symptoms using imaging when RT-PCR tests are limited.
• Evaluate multiple deep learning architectures on CT and chest X-ray datasets to identify robust COVID-19 classifiers.
• Incorporate explainability with LIME to interpret model decisions for radiologists.
• Provide comparative metrics (accuracy, precision, recall, F1) and confidence intervals to assess reliability across modalities.

Proposed method

• Use two datasets: 400 CT scan images (COVID-19 vs Non-COVID-19) and 400 chest X-ray images (COVID-19 vs Non-COVID-19).
• Apply transfer learning with eight pretrained CNNs: VGG16, InceptionResNetV2, ResNet50, DenseNet201, VGG19, MobileNetV2, NasNetMobile, ResNet15V2.
• Fine-tune by replacing the head; train with 80% data and test with 20% data; optimize with Adam and grid-searched hyperparameters (learning rate 0.001, 30 epochs, batch size 5).
• Evaluate using accuracy, precision, recall, and F1-score; report 95% CIs via Wilson score and Bayesian interval.
• Explain predictions with Local Interpretable Model-agnostic Explanations (LIME) to highlight top features influencing decisions.

Experimental results

Research questions

• RQ1 Which of the eight pretrained CNNs provides the best COVID-19 screening performance on CT and X-ray datasets?
• RQ2 How do model performance metrics (accuracy, precision, recall, F1) compare across modalities (CT vs X-ray) and models?
• RQ3 Can LIME explanations yield interpretable insights that align with radiological features for COVID-19 classification?
• RQ4 What is the stability of model training (training vs validation) and loss behavior across epochs for each model?

Key findings

• NasNetMobile achieves the highest test accuracy on CT scans (0.90) and on chest X-ray images (0.100? note: see below)
• On CT test data, NasNetMobile yields 90% accuracy and 0.90 precision, recall, and F1-score, with 8 misclassifications in the confusion matrix (80 test samples).
• On chest X-ray test data, NasNetMobile achieves 100% accuracy, precision, recall, and F1-score with zero misclassifications (80 test samples).
• MobileNetV2 performs strongest on the CT train data (accuracy 0.99) but NasNetMobile outperforms on the test set for CT scans.
• Across training runs, several models show stable training/validation performance (e.g., VGG16, InceptionResNetV2, NasNetMobile), while ResNet50 consistently underperforms.
• LIME is used to illustrate model explainability, demonstrating the feasibility of interpreting top features for trust in AI-based screening.