Article complete

How Is Open Source Transforming Medicine

AI policy is shaped by open source software and medical datasets are now open source, among other things.

A collage featuring a vintage doctor, a computer monitor, and data graphs on a bright yellow background.

Key takeaways

  • Open source software and medical datasets are reshaping AI policy in the healthcare sector.
  • Stanford's AIMI is creating a vast collection of free, AI-ready medical imaging datasets to promote research.
  • AI technologies are increasingly assisting doctors in diagnosing and predicting patient conditions.
  • Access to annotated medical datasets is crucial for the development of effective AI models in medicine.
  • Matthew Lungren advocates for medical data as a public good, accessible to researchers globally.

Takeaways drafted with AI from the article text.

Stanford's Center for Artificial Intelligence in Medicine and Imaging (AIMI) is expanding the world's largest free collection of AI-ready annotated medical imaging datasets to democratise artificial intelligence and medicine research.

Artificial intelligence has become an increasingly prevalent tool for interpreting medical pictures from tumour detection in mammography and brain scans to ultrasound video analysis of a pumping heart. Many AI-powered technologies currently assist human doctors in terms of precision. Some systems can do more than identify potential malignancies and fractures in the bones. These systems can also predict the patient's disease and offer recommendations.

However, AI technologies must be taught using expensive image datasets that human professionals have painstakingly annotated to save money. In part, research is supported by large organisations that don't necessarily share their data with the general public because datasets might cost millions of dollars to acquire or develop.

AIMI has amassed annotated photos from the Stanford University Medical Center, totalling more than 1 million. These datasets are available for free download by researchers, who can then use them to build AI models that suggest specific courses of action.

Creating an Ecosystem for Scientific Research

The goal is to develop a medical AI research environment that goes beyond picture analysis. People will investigate key clinical use cases beyond pixel data with correct datasets, including other associated multimodal data. Lungren expects the number of datasets at the centre, which already comprises over 1 million photos, to increase next year. The new platform will include two new datasets. Open-source data and common architectures can be used to create AI software in a box, which will, in turn, stimulate a new generation of crowd-sourced artificial intelligence research.

Democratising the Tools

Large firms may readily miss specialist areas like medical problems that affect certain populations, but academics will be able to investigate them by providing free data. In addition, researchers will be able to find biases in data and algorithms more easily with these diversified datasets. For example, according to research, some AI models are more accurate for particular populations than others because they were trained on patient data from a single area. Researchers will be better able to spot these problems if they have access to datasets from other populations.

Open-source software and artificial intelligence policy

The most extensively used machine learning technologies are powered by open-source software (OSS). Free technologies can—and should—play a larger role in AI policy discussions since they are available to everyone. According to a new Brookings Institution analysis, open-source must be more than just a policy footnote if the United States intends to lead global AI development.

Healthcare has the most to gain from AI's potential transformations of all industries. However, due to the high cost of training and acquiring datasets, much of this potential is still unused by large organizations. This platform can analyze more than one million photos and provides cloud-based processing capability by using up to nine datasets. 

To learn more, go to the Stanford AIMI Data Sharing Platform.

Published

Spotted an error? Tell our editors. Read our editorial standards and corrections.