MiniCovid-Unet: CT-Scan Lung Images Segmentation for COVID-19 Identification
Abstract: Detection and segmentation of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-COV2 or COVID-19) is a difficult task due the different kinds of shapes, sizes and positions of the injury. Medical institutions have vast challenges because there is an urgent need for efficient tools to improve the diagnosis of COVID-19 patients. Computer tomography images (CT) are necessary for medical specialists to diagnose the patient’s condition. Nevertheless, there is a lack of both in Medical Centers, mainly in rural areas. The manual analysis of CT images is time-consuming; in addition, most images have low contrast, and it is possible to find blood vessels in the background, so the difficulty of a suitable diagnosis increases. Nowadays, deep learning methods are an alternative method to perform the detection and segmentation task. In this work, we propose a novel light model to detect and identify COVID-19 using CT images: MiniCovid-Unet. It is an improved version of U-net; main differences reside on the decoder and encoder architecture, MiniCovid-Unet needs fewer convolution layers and filters because it focuses only on COVID-19 images. Also, as a result of employing fewer parameters, it can be trained in less time, and the resulting model is light enough to be downloaded to a mobile device. In this way, it is possible to have a quick and confident diagnosis in remote areas, where there exists an absence of internet connection and medical specialists.
| Main Authors: | , , , , |
|---|---|
| Format: | Digital revista |
| Language: | English |
| Published: |
Instituto Politécnico Nacional, Centro de Investigación en Computación
2024
|
| Online Access: | http://www.scielo.org.mx/scielo.php?script=sci_arttext&pid=S1405-55462024000100075 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract: Detection and segmentation of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-COV2 or COVID-19) is a difficult task due the different kinds of shapes, sizes and positions of the injury. Medical institutions have vast challenges because there is an urgent need for efficient tools to improve the diagnosis of COVID-19 patients. Computer tomography images (CT) are necessary for medical specialists to diagnose the patient’s condition. Nevertheless, there is a lack of both in Medical Centers, mainly in rural areas. The manual analysis of CT images is time-consuming; in addition, most images have low contrast, and it is possible to find blood vessels in the background, so the difficulty of a suitable diagnosis increases. Nowadays, deep learning methods are an alternative method to perform the detection and segmentation task. In this work, we propose a novel light model to detect and identify COVID-19 using CT images: MiniCovid-Unet. It is an improved version of U-net; main differences reside on the decoder and encoder architecture, MiniCovid-Unet needs fewer convolution layers and filters because it focuses only on COVID-19 images. Also, as a result of employing fewer parameters, it can be trained in less time, and the resulting model is light enough to be downloaded to a mobile device. In this way, it is possible to have a quick and confident diagnosis in remote areas, where there exists an absence of internet connection and medical specialists. |
|---|