Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Finding Covid-19 from Chest X-rays using Deep Learning on a Small Dataset (2004.02060v4)

Published 5 Apr 2020 in eess.IV, cs.CV, and cs.LG

Abstract: Testing for COVID-19 has been unable to keep up with the demand. Further, the false negative rate is projected to be as high as 30% and test results can take some time to obtain. X-ray machines are widely available and provide images for diagnosis quickly. This paper explores how useful chest X-ray images can be in diagnosing COVID-19 disease. We have obtained 122 chest X-rays of COVID-19 and over 4,000 chest X-rays of viral and bacterial pneumonia. A pretrained deep convolutional neural network has been tuned on 102 COVID-19 cases and 102 other pneumonia cases in a 10-fold cross validation. The results were all 102 COVID-19 cases were correctly classified and there were 8 false positives resulting in an AUC of 0.997. On a test set of 20 unseen COVID-19 cases all were correctly classified and more than 95% of 4171 other pneumonia examples were correctly classified. This study has flaws, most critically a lack of information about where in the disease process the COVID-19 cases were and the small data set size. More COVID-19 case images will enable a better answer to the question of how useful chest X-rays can be for diagnosing COVID-19 (so please send them).

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Lawrence O. Hall (6 papers)
  2. Rahul Paul (7 papers)
  3. Dmitry B. Goldgof (2 papers)
  4. Gregory M. Goldgof (5 papers)
Citations (247)

Summary

Diagnostic Utility of Deep Learning on Chest X-rays for COVID-19 Detection

The paper discusses the application of deep learning techniques to identify COVID-19 from chest X-ray (CXR) images, addressing the pressing need for rapid diagnostic alternatives due to constraints in traditional testing methods. This research leverages transfer learning from pre-trained deep convolutional neural networks (CNNs), specifically Resnet50 and VGG16, as well as a custom small CNN to discern COVID-19 from viral and bacterial pneumonia using a limited dataset of chest X-rays.

Methodology and Experimental Setup

The experimental setup involved a balanced dataset comprising 135 chest X-rays of COVID-19 patients and 320 X-rays from patients with either viral or bacterial pneumonia. Transfer learning was employed using Resnet50 and VGG16 models to accommodate the grayscale nature of X-rays, feeding the same image into all three RGB channels. The CNN architectures were adjusted by replacing the final layers with new layers suitable for the binary classification task, while retaining the feature extraction capabilities of the original models.

Data augmentation strategies, such as horizontal flipping, were used to slightly mitigate the limitations of a small dataset. A 10-fold cross-validation was conducted to gauge model performance, and a Snapshot ensemble approach was integrated to enhance predictive reliability. This ensemble involved averaging the prediction outcomes from a collection of 21 models trained on variations within the complete dataset.

Results

The application of deep learning models achieved a noteworthy overall accuracy of 89.2% when validated with 10-fold cross-validation on the balanced dataset, indicating a true positive rate of 80.39% for COVID-19 with a high specificity of 99% for pneumonia cases. On an independent test set, comprised of 33 COVID-19 cases and 208 viral and bacterial pneumonia cases, the ensemble model secured a true positive rate of 78.79%, along with a true negative rate of 93.12% and an AUC of 0.94.

Implications and Future Prospects

While this paper presents promising preliminary results, there are significant limitations, notably the small sample size, variable image resolution, and lack of detailed data on the disease stage within the COVID-19 cases. Such constraints impact generalization and highlight the requirement for a more expansive and diverse dataset. Additional data would allow for more robust model training and potentially enhance diagnostic power, particularly in distinguishing COVID-19 from various types of pneumonia under varying clinical scenarios.

Moreover, the discussion underscores the applicability of chest X-rays as a non-invasive, broadly accessible, and cost-effective diagnostic tool. With further advancements, models could serve as a supplemental clinical resource to quickly stratify patient risk and inform timely medical decisions in pandemic conditions where rapid diagnosis is critical.

Conclusions

The existing work demonstrates the feasibility of using deep learning with X-ray imaging for COVID-19 detection, but it underscores the necessity for more comprehensive data collection to overcome the current limitations. Plans for further investigation include exploring a broader dataset comprising high-resolution images, potentially augmented by pseudo-X-rays derived from CT scans, and developing models capable of more nuanced clinical insights. This approach, if advanced with adequate data, holds potential to enhance disease triage, optimize resource allocation in overburdened healthcare settings, and reduce dependency on conventional testing procedures in resource-limited environments.