Detecting SARS-CoV-2 From Chest X-Ray Using Artificial Intelligence.

Area of Science:

• Medical Imaging and Artificial Intelligence
• Infectious Disease Diagnostics
• Machine Learning in Healthcare

Background:

• Chest X-rays combined with Deep Convolutional Neural Network (CNN) methods show potential for COVID-19 detection.
• Existing methods face challenges with limited datasets, imbalanced data performance, and lack of confidence intervals.

Purpose of the Study:

• To evaluate the accuracy of six modified deep learning models for detecting Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection from chest X-rays.
• To address limitations of existing methods by testing models on balanced and imbalanced datasets with confidence intervals.

Main Methods:

• Six modified deep learning models (VGG16, InceptionResNetV2, ResNet50, MobileNetV2, ResNet101, VGG19) were trained and tested on chest X-ray datasets.
• Performance was evaluated using accuracy, precision, recall, and f-score on a small balanced dataset (Study One) and a larger imbalanced dataset (Study Two).
• Confidence intervals were calculated to assess the reliability of the results.

Main Results:

• VGG16 and MobileNetV2 achieved up to 100% accuracy with 95% confidence intervals on both datasets.
• InceptionResNetV2 and VGG19 demonstrated 97% accuracy across both datasets.
• Models like ResNet50 and ResNet101 improved from 70% to 93% accuracy when trained on the larger dataset.

Conclusions:

• Modified deep learning models, particularly VGG16 and MobileNetV2, offer a highly accurate and reliable method for COVID-19 detection using chest X-rays.
• The study highlights the effectiveness of deep learning in overcoming data limitations and improving diagnostic accuracy for infectious diseases.
• These findings present a promising, accessible alternative for identifying COVID-19 patients.