Doctor Analyzing X-Rays

General Computer Vision for Healthcare

Background

According to WHO, there will be a worldwide shortfall of 18 million health workers by 2030 [1]. While countries of all socioeconomic levels face this problem to varying degrees, the situation in developing countries is especially severe. Developed countries try to cope with this challenge by recruiting international medical staff from developing regions, but the resulting international migration of health workers even exacerbates workforce shortfalls in low- and lower-middle income countries [2].

Lack of doctors means that many people in need have no access to crucial health services. But also, those who have luck and can reach to the next hospital often get a treatment of poor quality or even an inaccurate diagnosis because doctors are often overstressed or there is no staff with the required domain knowledge available. In such situations, Machine Intelligence (MI) technologies might provide vital help.

Machine Learning (ML) systems are already proven to be able to make correct diagnoses in many medical fields, see, e.g., [3] for a review. While they cannot replace human doctors, intelligent diagnostic systems can support doctors in finding accurate diagnosis faster and allow them to spend more time with the patients looking for the best possible treatment.

Specialized ML systems, so-called Computer Vision, are especially effective in analyzing medical images, e.g., photos, X-ray images, CT and MRI scans. Such image analyses can be used by doctors for diagnosis of diseases or injuries. Thus, Computer Vision systems can support doctors, especially primary care physicians in developing countries who might not have enough special knowledge in specific medical fields.

However, current systems are very specialized. It means that they can identify only one or few diseases and are restricted to a specific set of data input, e.g., an X-ray image or a photo. In order to provide maximal added value for primary care physicians, especially in developing countries, an intelligent diagnostic computer vision system must be general, meaning that it must be capable of identifying a broad spectrum of diseases and dealing with many different types of input images.

In addition, due to the complexity and expenses involved, current systems are typically used only by high-end medical researchers and a very small number of medical practitioners. To make Computer Vision for Healthcare accessible for doctors in the developing world, one has to create a system that is open-source and free-of-charge.

Vision

A free-of-charge and easily accessible open-source general computer vision system that can identify a broad spectrum of diseases and can handle various sorts of input images (photos, X-ray images, CT and MRI scans etc.).

Expected Impact

This system will enable primary care physicians and other medical staff including those in developing countries to provide a more precise diagnoses for their patients. It will provide help towards resolving the problem of the shortfall of specialized medical staff in developing regions and lead to better healthcare service, less premature deaths, and higher quality of life for patients.

Project Phases

  • Phase 1: Research and Proof-of-Concept
    This phase is dedicated to research on possible MI-architectures that are capable of handling various sorts of input images and a broad spectrum of diseases. This phase serves as proof of feasibility for the planned General Computer Vision System for Healthcare and provides the scientific foundation for the whole project.

  • Phase 2: Building collaborations and collecting more data
    While there is enough publicly available data to perform Phase 1, in order to make a well working application, additional data must be collected and labeled so that MI-algorithms can learn from it. To collect this data, MI4People will collaborate with NPOs, hospitals, doctors, and governmental authorities.

  • Phase 3: Building Application
    This phase is focused on building a working prototype of the General Computer Vision System for Healthcare.

  • Phase 4: Field Study
    Using the results from Phase 3, a field study will be performed that measures the actual impact of MI-based general computer vision for healthcare, especially, in developing countries.

  • Phase 5: Release to Community
    If Phase 4 will show positive and impactful results, MI4People will make relevant software code, data, MI-models/tools, and any other intellectual property created during the project publicly available and will create a community of enthusiasts who will proceed to maintain and improve the application as an open-source project.

  • Possible Extensions
    If MI4People’s General Computer Vision System for Healthcare will have success, it might be reasonable to build upon it and create a system that not only provide diagnoses but also recommendations for treatment.

Current Status

Phase 1 is running.

Opportunities for Contribution

 

We are currently searching for Data Scientists who would be ready to contribute to the project as volunteers.

We also would be very grateful for any donations in cash or kind to support this initiative.