A Lightweight Deep Learning Model and Web Interface for COVID-19 Detection Using Chest X-Rays

• Published in: Traitement du signal 2024-02, Vol.41 (1), p.313-322
• Main Authors: Ainapure, Bharati Sanjay, Appasani, Bhargav, Schiopu, Adriana-Gabriela, Oproescu, Mihai, Bizon, Nicu
• Format: Article
• Language: English
• Subjects: Accuracy; Artificial neural networks; Chest; Coronaviruses; Cost effectiveness; COVID-19; Datasets; Deep learning; Feature extraction; Feature maps; Image filters; Image segmentation; Machine learning; Medical imaging; Neural networks; Polymerase chain reaction; Predictions; Public health; Real time; Viral diseases; X-rays
• ISSN: 0765-0019; 1958-5608
• DOI: 10.18280/ts.410126

COVID-19 is one of the deadly diseases that affected the global health system. It is difficult to diagnose COVID-19, as it shows the symptoms of the common cold. Therefore, effective screening techniques play a significant role in the timely detection of this disease. Existing techniques such as real-time reverse transcriptase-polymerase chain reaction (RT-PCR), require a considerable amount of time for processing, typically taking up to 48 hours to produce results. This delay can be detrimental, as the virus can spread rapidly during this waiting period. X-ray images are also used for this purpose due to their accessibility, speed, non-invasiveness, cost-effectiveness, ability to visualize lung tissues, and rapid deploy ability. This research proposes a convolutional neural network (CNN) to detect COIVD-19 based on chest X-ray images. The model's uniqueness lies in its ability to harness the power of convolutional layers for feature extraction without the need for complex segmentation techniques. The convolutional layers of the CNN filter slide across the input image, performing element-wise multiplication and accumulation to create feature maps. These maps highlight relevant patterns, edges, and textures present in the image. This can help in predicting the infection and its severity. With the proposed model an accuracy of 99% was achieved, and it attempts to balance computational efficiency and accuracy. Further, a web interface is developed so that users can use this model to obtain easy and accurate predictions. The proposed model aims to reduce the workload of healthcare workers and provide timely results to a patient so that further actions can be taken quickly.