Building a human-in-the-loop machine learning system to make medical image annotation as enjoyable as possible.
Extracting segmentations from medical images is labour- and time-intensive. The scarcity of labelled medical image data poses a significant challenge to the medical machine-learning community.
MIA allows medical experts to utilize the potential of machine-learning to quickly provide segmentation suggestions for medical images. These can be reviewed and verified and then be used to enhance existing models, creating a positive feedback loop of continuous improvement.
By providing a flexible way to manage datasets and an easy setup process, we allow researchers to focus on their actual task at hand.
MIA can be used by both individuals and large research teams. Its microservice architecture allows for secure deployment on-premise.
Medical image data requires special consideration for privacy. Data never has to leave the organisation when MIA is deployed on private clusters.
Create projects to organise your work and view images of datasets at a glance. Images can be annotated directly within the VISIAN editor.
Upload and track machine learning models, that you can use in jobs to generate segmentations quickly. Improve existing models with new ground truth data.
Correct auto-generated annotations easily and verify segmentations to mark them as ground truth. MIA allows you to track your annotation progress effortlessly.
is our system's central API, coordinating components, managing entities, storing metadata, handling image and annotation files, and queuing new jobs.
uses MLflow models to perform segmentation tasks, pulling jobs from a queue. Annotations are saved in the datastore and created as metadata entities in the hub.
enables MIA with model storage, tracking, and deployment for smooth data inference. It offers diverse model flavors, ensuring flexibility and compatibility across frameworks.
MIA integrates with AI4H Focus Group's assessment platform as a pre-annotation engine. Researchers can use machine learning to quickly create segmentation suggestions. These can be collaboratively reviewed using the VISIAN Editor, facilitating a multi-user collaboration in establishing ground-truth data for unbiased AI models.
MDC researchers contributed their expertise and testing to develop MIA alongside VISIAN. As a result, MIA can now streamline their workflows with seamless dataset management and enable efficient review and annotation of numerous images.
Visian is a tool that empowers researchers to annotate and analyse MRI scans faster than ever before. Its aim is to enhance the work of scientists in various fields with new approaches, features and use of machine learning to enable better, faster and more qualitative research.
On July 6, at the 20th Bachelorpodium, HPI's soon-to-be graduates presented innovative software solutions developed in teams over two semesters with external partners. Our focus was on MIA, addressing challenges in manually segmenting medical image data and demonstrating its facilitative features.
On June 8, the Hasso Plattner Institute's Digital Health Cluster hosted the German Federal Minister of Health, Karl Lauterbach, as a guest. The focus of Prof. Karl Lauterbach's visit was an exchange about current research initiatives in the field of Digital Health and Artificial Intelligence, which included MIA.