MD.ai: Supporting smarter medical AI research and development

About MD.ai

MD.ai assists in the entire artificial intelligence (AI) process from creating projects from medical reports, to annotating imaging data, and using the data directly in the app for training machine learning algorithms. MD.ai has developed an infrastructure for any radiologist or group to start learning AI and building their own projects.

Industries: Healthcare

Tell us your challenge. We're here to help.

Contact us

MD.ai leverages Google Cloud and Cloud Healthcare API to create annotated datasets and build algorithms for machine learning to bring better insights to medical providers.

Google Cloud results

  • Drives better patient care insights through smarter annotation and dataset analysis
  • Improves platform imaging, dataset analysis, and annotation abilities
  • Reduces time for project image management from hours to seconds
  • Creates a direct path to artificial intelligence integrations with Google Cloud via Jupyter Colab notebooks

Helped 1,300 teams process more than 30,000 images

Artificial intelligence (AI) holds great promise as the next big innovation for the future of medical imaging and patient care. AI is being used today as a valuable resource to produce insights on image quality control and identifying abnormalities for medical providers. From dataset construction to algorithm design, MD.ai specializes in bringing the infrastructure for AI-powered insights to hospitals and other healthcare partners.

For many of these providers, dataset development and annotation are among the biggest hurdles during the initial AI integration process. During the annotation process, evaluators typically markup files to teach models what they should focus on or ignore. With medical applications, this could include highlighting tumors on patient scans to help models more easily identify potential risk areas. For medical imaging, evaluators usually need to have special training. Difficult entities, such as pneumonia, require radiologists and even thoracic radiology specialists to get high-quality labels. Simpler annotation, such as organ segmentation can often be done by technologists.

Considering the number of resources needed to create strong models, even large companies regularly turn to public datasets for training algorithms and crude open-source applications for annotation. However, these datasets often have issues with mislabeled or disorganized labels that can hinder proper model calibration and lead to poor data integrity. For healthcare providers, low-quality models run the risk of producing false positives or even worse, false negatives, and limit the potential benefits offered by machine learning. The open-source annotation software often doesn’t have a GUI interface and may be limited in scope.

“For AI development, annotation is the biggest hurdle faced by customers. Google Cloud provides our partner hospitals with a reliable foundation to integrate quality patient datasets at scale.”

Anouk Stein, M.D., Board Certified Radiologist, MD.ai

To help solve this challenge, MD.ai leveraged Google Cloud to power its Annotator solution, which provides scalable and easily approachable annotation features for dataset development. Data needs to be properly classified for AI development and from the platform-wide integration of Google Cloud and its services, MD.ai has already benefited from Google’s approach towards making data organization smarter. With its strong annotation abilities, the MD.ai service helps healthcare providers create higher-quality medical insights to help improve patient care.

“For AI development, annotation is the biggest hurdle faced by customers,” says Anouk Stein, M.D., Board Certified Radiologist, MD.ai. “Google Cloud provides our partner hospitals with a reliable foundation to integrate quality patient datasets at scale.”

Cloud-powered annotation

Multiple Google Cloud components power the Annotator backend. Google Kubernetes Engine (GKE) forms the base for the application, as it helps with usage-based scaling and helps store project data on a dedicated cluster. Annotator is also fully integrated with Cloud Healthcare API, which provides a suite of healthcare data-focused features. These include implementing HL7 and DICOM protocols natively, along with providing support for industry security standards and HIPAA compliance. Annotator also benefits from the Google Cloud architecture, helping to ensure maximum uptime and remote file accessibility.

“Google Cloud and Cloud Healthcare API gave us an invaluable head start when creating Annotator. By not having to invest time building out the initial groundwork, we could focus more on larger platform development that enables healthcare providers to improve patient outcomes.”

George Shih, M.D., Co-founder, MD.ai

The Annotator tool comes with a full set of features designed to meet the organization-wide needs of businesses and hospitals interested in developing AI. In the application, customers can search for images based on customized tags, as well as markup and comment on files to highlight or exclude specific regions. Annotations are also exportable in formats including JSON and DICOM SR, which are used for medical text files and images. The native support built within Annotator helps encourage intuitive collaboration without spending time manually converting documents.

Because MD.ai built and optimizes Annotator for Chrome, customers gain additional productivity. Legacy annotation solutions required medical researchers to download local applications and massive file batches that could often include thousands of patient scans. By making Annotator a cloud-native application, doctors and researchers don’t need anything except a web browser for annotation, which makes work easier and removes complexity for both customers and MD.ai.

“Google Cloud and Cloud Healthcare API gave us an invaluable head start when creating Annotator,” says George Shih, M.D., Co-founder of MD.ai. “By not having to invest time building out the initial groundwork, we could focus more on larger platform development that enables healthcare providers to improve patient outcomes.”

Annotating in action

For MD.ai, Google and Google Cloud were vital contributors to the success of the Radiological Society of North America (RSNA) Pneumonia Detection Challenge. As part of the company-sponsored testing event, MD.ai tasked teams with building a machine learning-powered algorithm that predicted the likelihood of pneumonia in patients by analyzing sample radiograph images for lung opacity indicators.

During the process, teams used Annotator and a 30,000-image dataset from MD.ai that was split between exams with potential pneumonia classifications, pneumonia-free labeling, and entirely unmarked images. Additionally, images were classified with five separate labels to evaluate the probability of an accurate pneumonia classification.

With more than 1,300 competing teams spread throughout the world, MD.ai and Annotator relied heavily on Google and Google Cloud to help ensure that the competition ran smoothly. Along with the ease of use that Google Cloud offered for team cloud file access, GKE supported the service and team needs for project creation, data storage, importing, and exporting datasets, and algorithm creation.

“Over 3,000 research participants joined in the RSNA Challenge. The success of the challenge wouldn’t have been possible without Google and we’re confident this will carry over into future machine-learning-powered research discoveries.”

Anouk Stein, M.D., Board Certified Radiologist, MD.ai

MD.ai has already drawn valuable insights that will be applied for other development and future competitions and is working on a new annotation initiative that carries over the competition’s format to bone imaging and detecting fractures and other findings like orthopedic hardware.

Internally, MD.ai and its Annotator optimized for Chrome produced additional performance gains for the tool, which helped support features including interpolated image annotation. Colaboratory support, a free Jupyter notebook environment, is also integrated into Annotator to smoothly transition between annotation and model development.

“Over 3,000 research participants joined in the RSNA Challenge,” says Dr. Stein. “The success of the challenge wouldn’t have been possible without Google Cloud and we’re confident this will carry over into future machine-learning-powered research discoveries. We hope others will further test and modify the top models to achieve clinically useful results.”

Classifying success

When delivering patient care, doctors can benefit from resources that improve decision-making and speed. With Google Cloud AI capabilities, MD.ai provides an easy and seamless interface for creating AI projects from reports, annotating data, and building algorithms for medical machine learning. The MD.ai Annotator can be used by teams of physicians or other trained personnel to annotate imaging data. It’s essential for radiologists and other physicians to be involved in creating AI projects because they understand current diagnostic challenges and are in the best position to determine the future of medical AI.

“AI can be a powerful tool, but it needs high-quality data,” says Dr. Shih. “Annotator reliably produces this for our customers, thanks to the foundation we built with Google Cloud.”

Tell us your challenge. We're here to help.

Contact us

About MD.ai

MD.ai assists in the entire artificial intelligence (AI) process from creating projects from medical reports, to annotating imaging data, and using the data directly in the app for training machine learning algorithms. MD.ai has developed an infrastructure for any radiologist or group to start learning AI and building their own projects.

Industries: Healthcare