DeepHealth: A Deep Learning Solution for the COVID-19 crisis
Winner of the COVID-19 AI Challenge from Saturdays.AI
Disclaimer:
This blog post on automatic COVID-19 detection is for educational purposes only. It is not meant to be a reliable, highly accurate COVID-19 diagnosis system, nor has it been professionally or academically vetted.
Motivation
With the World Health Organization (WHO) declaring coronavirus (COVID-19) a global pandemic, governments have issued guidance for members of the community to practice social distancing, while companies have enforced work from home policies in an effort to flatten the curve of viral infections across the population.
One of the takeaways is that in many countries, like Spain, it is very likely there are 100x infected when comparing reality with official records.
- Our colleague Carlos Bort, wrote this article based on data from Madrid (Spain) to estimate what could be the actual number of infected a few weeks ago.
- A team from Celera created their TheOpenVentilator project to help democratize respirators in a cheap and scalable way.
- Christian Tutivén, Ángel Encalada y Andrés Torres created a working group in order to help Ecuador’s public health system.
- Eduardo H. Ramirez is co-leader of an initiative in Mexico to track data from the country in order to measure changes in the spread of the virus in real time.
Motivated by the work of our colleagues, we decided to launch this challenge with the aim of finding a solution to the crisis of Covid-19.
We proposed an open-source Data Science solution to tackle the Coronavirus crisis using Machine learning, Data Visualization, Deep learning, etc. With the reward of 1k€ cash + 1k$ AWS credits + other perks to the best original solution to tackle the Covid19 problem.
When we launched the challenge we didn’t imagine the reception it was going to get.
Which were the finalists?
More than 20 teams from up to 6 different countries signed up. And these were the 4 finalists selected:
- Team 1: Deephealth — Training of an image classifier of thoracic plaques (X-ray), which differentiates between non-pneumonia, bacterial pneumonia, viral pneumonia (the case at hand). Its focus is to assist in early detection using readily available means as an alternative to PCR which can be complicated/costly to obtain.
- Team 2: (GeoDB) — Use of the SIR model to predict the growth of those infected, recovered and killed by the disease by ingesting JHU data to make estimates for different countries and communities.
- Team 3: (Appoyo) — A chatbot is proposed with information for citizens on different topics such as employment, health or solitude during confinement. This chatbot will have options for interaction through the web, whatsapp or telephone.
- Team 4: (Covisual19 System) — The solution consists of a dashboard deployed in Flask that obtains official data from the Mexican government website and makes state-level predictions using the SIR model. It also has an interface that allows navigating a map with pins to separate by the different states.
🏆 Winner: Deephealth
Author: Miguel Ángel Salinas
The risk of pneumonia is immense for many, especially in developing nations where billions face energy poverty and rely on polluting forms of energy. The WHO estimates that over 4 million premature deaths occur annually from household air pollution-related diseases including pneumonia. Over 150 million people get infected with pneumonia on an annual basis especially children under 5 years old.
In such regions, the problem can be further aggravated due to the dearth of medical resources and personnel. For example, in Africa’s 57 nations, a gap of 2.3 million doctors and nurses exists. For these populations, accurate and fast diagnosis means everything. It can guarantee timely access to treatment and save much needed time and money for those already experiencing poverty.
Today, pneumonia screening is a manual process that requires a trained clinician to examine and evaluate radiographic plaques for each affected patient.
The need for a complete and automated method to detect pneumonia, the main cause of mortality of Covid-19 virus, makes it necessary to find other detection methods to help detect pneumonias caused by Covid-19.
Project: Build an algorithm to automatically identify whether a patient is suffering from pneumonia or not by looking at chest X-ray images. The algorithm had to be extremely accurate because lives of people is at stake.
- DeepHealth Model: thingtrack/deephealth-pneumonia-model
https://github.com/thingtrack/deephealth-pneumonia-model.git - DeepHealth Backend: thingtrack/deephealth-pneumonia-api
https://github.com/thingtrack/deephealth-pneumonia-api.git - DeepHealth Frontend: thingtrack/deephealth-pneumonia-ui
https://github.com/thingtrack/deephealth-pneumonia-ui.git
Technical documentation and scalable architecture based on system microservices. Deep Health is not only a predictive model, it is a scalable web application that can be easily used by medical personnel from a web browser under Kubernetes.
Data
DeepHealth gather X-ray data from healthy patients, patients with viral pneumonia, patients with bacterial pneumonia, and COVID patients. DeepHealth build a Inception V3 model, implementation of transfer learning, and testing of how well the model can classify in three different scenarios obtaining 83% accuracy in training:
- COVID vs Normal
- Viral Pneumonia vs Bacterial Pneunomia vs Normal
- COVID vs Viral Pneunomia vs Bacterial Pneunomia vs Normal
Why Chest X-ray?
A Chest X-ray is a type of imaging study for respiratory issues. In severe COVID-19 cases, chest X-rays show signs of pneumonia, and these X-rays can help better understand the health of a patient’s lungs.
In contrast, a CT scan is a more expensive and specialized imaging study that uses X-rays for 3D imaging of the chest. Although CT forms a better picture for diagnostics X-rays are more widespread.
Wait, what’s Transfer Learning? 🎯
It’s a deep learning concept, you grab an existing saved model that’s already spent time training on a lot of data and apply the features it has learned from it’s experienced to your new problem. (Kinda like if you could just borrow someone’s experiences and expertise with playing classical piano, then jazz it up.)
You can tweak things and alter them to fit your situation, like only using part of the trained model and not using the other layers. Transfer learning lets you leverage past knowledge of related tasks to solve the current problem.
Try the DeepHealth Tool:
Once the Neural Network model has been chosen and the model has been trained. Deep Health allows access to this model online, through a browser with a simple Internet connection. This extremely simple tool allows the medical specialist to simply choose the X-ray image of the patient’s thorax that he or she wants to diagnose and let the model’s Artificial Intelligence classify the patient within one of the 3 possible ranges of Pneumonia.
- Deep Health Pneuma 👉🏻 http://pneuma.deephealth.thingtrack.com/
Conclusion
We evaluated all proposals and finally chose DeepHealth as the most impactful and best documented solution.
COVID-19 tests are currently hard to come by — there are simply not enough of them and they cannot be manufactured fast enough, which is causing panic.
When there’s panic, there are nefarious people looking to take advantage of others, namely by selling fake COVID-19 test kits after finding victims on social media platforms and chat applications.
Given that there are limited COVID-19 testing kits, we need to rely on other diagnosis measures.
Acknowledgements
Data: https://data.mendeley.com/datasets/rscbjbr9sj/2
License: CC BY 4.0
Citation: http://www.cell.com/cell/fulltext/S0092-8674(18)30154-5
About Saturdays.AI
Saturdays.AI is an impact-focused organization on a mission to empower diverse individuals to learn Artificial Intelligence in a collaborative and project-based way, beyond the conventional path.
