Overview
The global pandemic COVID-19 is spreading rapidly across the world.
Big or small, every organization is trying hard to find strategies to fight the COVID-19 outbreak. Just like many of you, even we, at Cigniti, are lending our support to governments, communities, civil society, and frontline health workers globally, and are exploring opportunities to bring multiple initiatives to scale.
Data science. Artificial Intelligence. Machine learning. Our weapons to fight the outbreak.
We are lending our expertise and experience in this field to address the challenges.
Cigniti welcomes the support of doctors, machine learning practitioners, developers, health practitioners, data science students, or anyone interested in contributing to these initiatives.
Brief Summary
COVID-19’s rate of transmission depends on the timely detection of the carrier and the immediate implementation of interventions. Even though the CT scan is more sensitive to COVID Pneumonia, our approach considered Chest X-rays, for a possible preliminary classification, due to its prevalent usage as a primary diagnostic test.
As linear dimension reduction techniques (like SVD) failed to classify the COVID Pneumonia from the Normal and CAP X-Rays, we have used Uniform Manifold Approximation and Projection (UMAP) to do the trick. UMAP uses manifold learning techniques and ideas from topological data analysis for dimension reduction.
Using Supervised UMAP, we have been able to separate the 3 classes in the dataset. Our current dataset is limited, and hence not representative of the larger Covid-19 positive population. We hope to remedy that and use UMAP as a feature extraction technique for a classifier in the coming few weeks.
The Roadmap
Brief Summary
Background:
Basic research done so far indicates that the apps and services that are targeted at frontline workers include:
- Emergency Information Broadcast; Contact details and announcements
- General Purpose Information Apps
- Traceability Apps; though these aren’t specific for frontline workers.
Our POV:
We have seen a definite need for “out of the box” solutions that can help with triaging and assessment, of who needs to get the tests done on a priority.
Develop a mobile application that can help understand if the symptoms are indicative of COVID-19 with a high level of accuracy. This will prevent the healthcare system from getting overwhelmed as well as get the testing done for patients at high risk.
We, at Cigniti, have looked at 2 variants of this triage.
STREAM 1: Spectral analysis of cough sample to predict COVID-19 possibility
STREAM 2: End-to-end triage, along with specific “Call For Action”, focussed on frontline workers
The Roadmap
Build a white-labelled app (+ backend dashboard) for an organization that has people working on a daily basis during the pandemic. This could be the police force, sanitation workers, retail outlets, technicians and staff in hospitals, etc
- All the staff has to do is answer a set of standard triage questions on a daily basis. The intent is to keep it as less subjective as possible. Few ways that we intend to do this:
- Instead of people answering a question around breathlessness, we ask them to record their voice with numbers 1 to 10. The app uses the audio to detect breathlessness.
- For people experiencing cough, the app will have a provision to record the cough so that audio analysis to understand the severity.
- All the data is collected on a daily basis (or any number of times in a day). This coupled with the location information (which can be mapped to red zone data) will provide the longitudinal data needed for the analysis.
- Any anomalies that are detected (example – mild temperature + breathlessness + loss of smell over a period of 3 days) will be flagged as Risk
- The medical helpline team within the organisation will have access to the data being collected and will also be able to view the users with higher risk on top
Through this process, we would also like to collect as much data as possible in the form of audio samples of cough and other symptoms that people are dealing with.
With that as an intention we are also looking at launching the app for the general public – anyone who is in self-isolation, anyone who is intending to digitally record their recovery process can use the app and track the progress.
Our collective agenda should be to collect as varied and as much data as possible (subjective and objective) so that we can let the clinical and tech community analyze it for a better understanding of this pandemic.
Brief Summary
Should the lockdown be extended? Should the restrictions be relaxed? How effective have the non-pharma interventions been, in controlling the outbreak? What is the rate of infection spread? What needs to happen in order to lift the lockdown (from an epidemiology standpoint)?
Questions like these made us pursue a framework where we have used effective reproduction number to give us insights on whether the non-pharmaceutical interventions should be tightened or relaxed.
The Roadmap
COVID-19 RT Tracker in INDIA:
Want to collaborate?
We believe that COMMUNICATION. COLLABORATION. COOPERATION can accelerate the control of COVID-19. It could be sharing your resources, volunteering a few hours every day, forwarding these initiatives to the larger community, or through sharing any of the below.
Big or small; everything counts & adds!
DataSets
Domain Expertise
Networking
Get Involved
We are all the catalysts in this big initiative. Let us collate and work together to make these initiatives a reality.
We are open to peer review and constructive feedback.
- Feedback Welcome: For providing your constructive feedback or collaboration interest, please feel free to write to us –marketing@cigniti.com