EnglishPortugueseSpanish

Assistance to health professionals in the diagnosis of COVID-19 by analyzing lung tomography through neural networks

Project summary

The first step in the treatment and isolation of patients with suspected Covid-19 is the diagnosis of the disease, so that necessary measures can be taken. The new disease left many health professionals unprepared to make an accurate diagnosis based only on existing techniques for other diseases. In developing and populous countries like Brazil, laboratory tests are used in the diagnosis of the disease but they end quickly, requiring a quick replacement, which usually does not happen. A solution to this problem may be the use of more accurate, affordable, and efficient Covid-19 diagnostic software. As the lung is the target of Covid-19 infection, we believe that data on the pulmonary condition of patients is the most useful way to define the diagnosis. Clearer, more understandable and easier to obtain data from organs such as the lung are tomographic images. We developed a program using Artificial Intelligence (AI) to detect Covid-19-related patterns in patient lung CT scans together with a convolutional neural network (CNN's) architecture basis, as they are very useful in analyzing spatial patterns in images, that determines with reliable accuracy whether a lung CT belongs to a healthy person, or to a person with Covid-19, or to a person with another lung disease. This was done using a public multiclass database of CT scans of patients diagnosed with Covid-19, CT scans of healthy patients, and CT scans of people with some other lung disease. The project consisted of, after creating the initial program, carrying out the AI ​​training. This training consisted of changing the architecture, parameters and meta-parameters, processing and organizing the data, applying different AI techniques resulting in a gradual improvement in the performance of the neural network. As it is an AI training process, different information was sought that related the AI ​​and medicine areas so that, during this process, there was an improvement in the program's performance. The data used to train and assess AI was obtained from [2] and contains 4173 CT scans from 210 different patients: 758 CT scans from healthy patients, 2168 CT scans from patients with SARS-Cov-2 and 1247 CT scans from patients with other lung problems. All these data come from the Hospital do Servants Estadual de São Paulo and the Hospital Metropolitano da Lapa, also in São Paulo, and the images are anonymous. At the current stage of software development, correct diagnoses were achieved with 84% accuracy in diagnosing people with Covid-19, or healthy, or with some other lung disease. These results were compared to theoretical results (diagnostic accuracy of patients with Covid-19 obtained by physicians with symptom analysis in conjunction with laboratory tests as in [5]) which ensure 70% to 80% accuracy in diagnosis.

Students

Caio Petroncini
Bernardo Kretzer da Silva Venzon Orlandi

Guidance counsellors

Cristiane Maria Alves Pissarra Fernandes
Leonardo Gomes Oliveira

Institution

Unisociesc International School (EIU) – Florianópolis
Florinópolis /
  SC -
  Brazil

Popular vote*

Did you like it? So vote and share now:

71+

Leave your comment

What did you think of this project? Participate by leaving your comment below:

Subscribe now
Notify of
guest
6 Comments
Oldest
Newest Most Voted
Inline feedback
View all comments
Alexandre Morais da Rosa
Alexandre Morais da Rosa
1 year ago

Very cool! Congratulations

3+
Cristiane Maria Veiga Amirim
Cristiane Maria Veiga Amirim
1 year ago

Excellent job. An important analysis for our world moment‼️ Gratitude

5+
Arthur
Arthur
1 year ago

I was very impressed with the idea. Congratulations!

1+
Hero Welter
Hero Welter
1 year ago

Congratulations on the project!

1+
Emerson Venzon Orlandi
Emerson Venzon Orlandi
1 year ago

Excelente trabalho.
Congratulations Caio and my son Bernardo.

1+
Niara Orlando
Niara Orlando
1 year ago

Congratulations to the beautiful work!!! Exciting ❤️

0

Popular vote*

Did you like it? So vote and share now:

71+

Students

Caio Petroncini
Bernardo Kretzer da Silva Venzon Orlandi

Guidance counsellors

Cristiane Maria Alves Pissarra Fernandes
Leonardo Gomes Oliveira

Institution

Escola Internacional Unisociesc (EIU) – Florianópolis
  SC –
  Brasil

Educational Incentives and Awards Offered

Research Areas

Get to know the projects that are competing in all of Mostratec Virtual's research areas:

Cellular and Molecular Biology and Microbiology
Biochemistry and Chemistry
Environmental Sciences
Animal and Plant Sciences
Computer Science
Planetary and Earth Sciences and Mathematics and Physics
Health Sciences
Education and Humanities
Environmental and Sanitary Engineering
Engineering and Materials
Electrical engineering
Electronic Engineering
Mechanical Engineering
History and Social Sciences
6
0
Click to leave your comment and participate!x
()
x

Popular Jury Awards

For the People's Jury awards, only single votes (one per user) and that are cast between 0/00 and 26:10 on 23/59 will be considered valid.

Entrepreneur Profile Test

Fill in your details below to perform your test: