RedNeumon: Convolutional Neural Network Model to Detect COVID-19 Infection
Study Details
Study Description
Brief Summary
The purpose of this study is to design a Convolutional Neural Network (CNN) and apply an attention model to help differentiate pneumonia due to SARS-CoV-2, pneumonia due to other causes and normal chest radiographs in clinical practice using a bank of digital chest images from a high complexity health facility in Cali, Colombia.
Condition or Disease | Intervention/Treatment | Phase |
---|---|---|
|
Detailed Description
This work focuses on developing an optimized transformer model to classify x-ray images into normal, abnormal with pneumonia caused by SARS-CoV-2, and abnormal with pneumonia due to other sources. Using a private dataset, the developed model can detect all three cases with high accuracy. Other datasets were used to confirm the model's accuracy, which provided positive results. The model is based on self-attention mechanisms and convolutional layers, which have been shown to improve performance when combined. The proposed model was tested on various datasets and reached satisfactory results in terms of accuracy and speed.
Moreover, the model can easily apply to different tasks, as it can quickly be retrained with new datasets. The proposed model combines Convolutional Neural Networks (CNNs) and optimized Attention Mechanisms. Modifications to the attention mechanisms were made to accelerate the algorithm by implementing a CNN for the linear projection. We obtained an accuracy of 80.1% for the model and a precision of 88%, 79.2%, and 72% for pneumonia caused by SARS-CoV-2, pneumonia due to other sources, and normal x-ray, respectively.
Study Design
Arms and Interventions
Arm | Intervention/Treatment |
---|---|
Normar chest radiographs X-rays without alterations in the lung parenchyma |
Other: Categorization of chest xrays images
Use of Convolutional Neural Network Model to categorize chest xrays images in each group.
|
COVID-19 chest radiographs X-rays belonging to patients with a diagnosis of COVID-19 confirmed by positive Reverse Transcriptase polymerase chain reaction (RT-PCR) and/or presence of antibodies to COVID-19 and/or positive COVID-19 viral antigen. |
Other: Categorization of chest xrays images
Use of Convolutional Neural Network Model to categorize chest xrays images in each group.
|
Other pneumonia chest radiographs X-rays belonging to patients with a diagnosis of pneumonia other than COVID-19 |
Other: Categorization of chest xrays images
Use of Convolutional Neural Network Model to categorize chest xrays images in each group.
|
Outcome Measures
Primary Outcome Measures
- COVID19 pneumonia chest radiograph identifyed [month 8]
Development and determination of the predictive capacity of a Convolutional Neural Network model to detect viral pneumonia in chest radiographs of adult patients with acute respiratory disease secondary to SARS-COV-2 infection.
Eligibility Criteria
Criteria
Inclusion Criteria:
-
Chest radiographs from patients without COVID-19 or other pneumonia took before the pandemic start date (January 2020)
-
Chest radiographs from patients with COVID-19 confirmed by positive Reverse Transcriptase polymerase chain reaction (RT-PCR) and/or presence of antibodies to COVID-19 and/or positive COVID-19 viral antigen.
-
Chest radiographs from patients without COVID-19 confirmed by a negative Reverse Transcriptase polymerase chain reaction (RT-PCR) and other pneumonia diagnoses taken before the pandemic start date (January 2020)
Exclusion Criteria:
- N/A
Contacts and Locations
Locations
Site | City | State | Country | Postal Code | |
---|---|---|---|---|---|
1 | Fundacion Valle del Lili | Cali | Valle Del Cauca | Colombia | 760001 |
Sponsors and Collaborators
- Fundacion Clinica Valle del Lili
- Universidad Autonoma de Occidente
Investigators
- Principal Investigator: Liliana Fernandez, M.D, Fundacion Clinica Valle del Lili
Study Documents (Full-Text)
None provided.More Information
Publications
None provided.- 1805