Artificial intelligence is revolutionizing medical screening, diagnosis, and drug discovery.
Datamint empowers all medical researchers to conduct cutting edge AI research, without a Computer Science PhD or a dedicated AI lab.
Medical AI research best practices, data management, annotation support, experiment tracking and reporting, and cutting edge data training techniques like Active Learning, all in one platform.
Research Software for the AI Curious
Datamint is an AI Lab in a Box
We give you all the tools needed for a successful medical AI research project.
Cutting edge research practices
Data retention, auditing and tracking, experiment replication. You are smart enough to figure all this out and set it up yourself. Do some research instead.
Tools like Active Learning support let you get your research done in 20% of the time of a traditional project, often without recruiting external readers.
Reporting and Tracking
Stay on top of how your project is progressing with model and experiment analytics that let you get a top-level view or dive deep into the weeds.
Try out every model version on new data. Impress your friends and make your colleagues jealous by adding a new image and seeing your research do its magic.
Help when you need it
Maybe you have a nephew with a PhD in medical machine learning, or maybe you need some help. Either way, we've got you covered with expert-level tools for your team, or expert researchers ready to assist.
Active Learning
Using active learning you can speed up training of models and achieve your research goals faster.
Automatic iterations mean you don't need to rely on programmers. Reduced annotation demands mean your project can proceed without waiting for expensive, external readers. You don't need your own compute facilities or programmers.
All you need is your idea and your passion for results.
If your fracture detection model was 95% accurate on arms but 5% accurate on legs, what kind of images would you annotate to improve the model? That is the essence of active learning.
Traditional Machine Learning
-
Annotate hundreds of images
-
Train a model
-
Hope for the best
Active Machine Learning
-
Annotate only a few images
-
Automatically train a model
-
Run hundreds of predictions, identify the worst
-
An expert is asked to annotate the 5 worst predictions
-
A new model is trained with the corrected predictions. Repeat until desired results achieved.
Features
01.
Data Cleaning and Management
-
Get your data into the system easily, with optional and flexible anonymization.
-
Clean and preprocess your data faster than ever before. Crop images and video one by one or in bulk, delete video frames, mask burned in PII in minutes.
-
Flexible and intuitive organization and searching of your complete data corpus.
-
Full support for all DICOM based modalities as well as natural video for endoscopy.
02.
AI Enabled Annotation
-
Streamline and manage data annotation and labelling with a few clicks.
-
Accelerate labelling with built-in AI assisted segmentation.
-
Invite external readers to annotate your images, and be proud of the streamlined experience they encounter.
-
Conduct annotation in parallel with AI training, knowing your researchers are always using the most up-to-date datasets.
03.
Experiment Tracking and Management
-
Seamless integration with frameworks like Pytorch allow your programmers to access the latest data with no effort.
-
Automatic tracking of experiment progress and detailed comparisons across experiments.
-
Every experiment is tied to a version of the dataset that existed when the model was trained, ensuring everything can be replicated.
04.
Model and artifact management
-
Your experimental results - each model, its accuracy, architecture, hyper parameters, and the road it took to get to where it is, can be automatically saved right from your python code and made accessible for you and your team. It's a zero-effort lab notebook that never gets lost.
-
Later, use your models as a launchpoint for future projects, or apply them against new datasets just to see what happens.
05.
Active Learning
-
Stop choosing images to annotate at random, like a caveman. Cutting edge algorithms optimize query-oracle interactions to achieve targeted levels of accuracy with a fraction of annotated images.
-
Datamint encodes a decision tree of best practices for selection of images to annotate, while letting you override when necessary.
-
Shrink your project time from years to months.
Join Our Mission.
Datamint is in closed beta. We are seeking passionate medical researchers and clinicians who want to expand their research into AI.
AI is going to change medical practice forever. You're itching to be part of that -- let Datamint help.
Let's go.
Reach out to ask us to discuss how your project can start using Datamint
About Us
Team and Mission
We are a technology company located in Edmonton Alberta, founded by AI experts, seasoned entrepreneurs, and advised by senior AI leaders with expertise in medical imaging AI research. Our belief is that AI will revolutionize healthcare practice in ways we cannot yet imagine - but that AI researchers are the engine that will drive the revolution forward. We are supercharging their efforts by providing platforms and tools that will be at the core of research teams' daily work.