Feasibility

Feasibility

Overview: Can It Work?

Your first reaction will likely be skeptical, because an app like this would not have been feasible even five years ago. Caredemic is now possible because:

  • There are precedents of similar projects: Details are below.
  • Pleasant Solutions has the relevant horsepower with departments for advanced software development, machine learning, electrical engineering, and even audio experts. Above all, we have the willingness to try without delay and without the goal of profiting from it. Coincidentally, we happened to have been working on machine learning using low-cost device microphones before COVID-19 pandemic.
  • People want to help. We believe we can attract 10,000 volunteer subjects or more, because testing can be performed without discomfort, at home, in only a few minutes. People want to feel like they are contributing to a solution and not just sitting, bored in isolation, waiting for things to unfold.
  • Phone hardware has come a long way in the past 5 years.
  • Machine learning has matured in the past 5 years. Our understanding has expanded exponentially, as has the global pool of experts. Machine learning has even modeled the interaction of atoms to predict drug side effects. This project is relatively straightforward compared to that, albeit on a much shorter timeline.
  • We aren’t doing it alone. We are reaching out to dozens of machine learning teams globally to participate (see contributors).
  • Medical information surrounding COVID-19 is still limited, but are supportive of the feasibility and value of this project. See references for more.

What More Standard Process Is This An Alternative To?

We are not trying to replace or compete with laboratory tests for diagnosis.

We are trying to “compete with”:

  1. Screening and sadly even diagnosis by medical staff, doctors, or the patient themselves, based on their observations or basic questionnaires. Capacity issues are leaving people with no choice, and we understand that, but this method is being used despite being very limited and poor accuracy at early stages, and largely infeasible when pre-symptomatic. We are working to develop a tool to accurately detect COVID-19 signs before individuals or doctors can identify any symptoms, even with a stethoscope, such as with “silent pneumonia”.
  2. “Do nothing”. Sadly, this is a year 2020 reality in even 1st world countries, due to capacity issues. Caredemic can assist without extra burden to either supply chains or medical institutions.
  3. Unreliable self-diagnosis by citizens researching online. Absent the ability to reach healthcare professionals, they will make judgements on their own. It is unavoidable, but we can give them a way to improve the accuracy and be aware of more thoughtful conclusions.

What Is the Chance Of Success?

For some usages, the probability of success is >90%. For others, it may be 20% but increasing daily.

In short, it is better than many other avenues of research currently being explored, and its impact could be more immediate. These odds are high enough that the potential must be explored. Within weeks of full-throttle effort, feasibility of each usage case could be more accurately pinned down, but there is only one way to quickly find out.

Previous Mobile-Based Medical Projects That Suggest Feasibility

In short, we have seen phone apps without equipment with sufficient accuracy to be approved for use in various parts of the word for diagnosis of heart conditions, diagnosis of COPD and pneumonia via coughing, among other uses.

Specific examples:

  • Cell phone motion sensors used to predict cardiopulmonary status with a high degree of accuracy (motion sensors used to measure walking patterns which in turn speak to oxygen levels, which in turn resulted in accurate categorization of cardiopulmonary status). 20 subjects. It appears they used special hardware to augment the phone with oxygen data or at least compare data to, though. Source: Feb 2016, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4744879/
  • In Australia, doctors have demonstrated good results using mobile phones to recognize types of cough, with no additional hardware required. https://respiratory-research.biomedcentral.com/articles/10.1186/s12931-019-1046-6
  • Carnegie Mellon Unviersity has made progress on an app that detects Covid by coughing and talking. Our platform would go several steps further but incorporate their progress. https://www.cultofmac.com/698029/researchers-detect-covid-19-listening-voice/
  • Less directly relevant, but a reminder of what machine learning is capable of: machine learning has even be used to model atoms interacting to predict drug side effects. It feels like our task at hand is easier than that, but the timeline is shorter.

COVID-19 Information That Suggest This is Feasible

See References section.