←
AI-enabled healthcare diagnostic-assist tools and patient outcome predictors
Industry
Healthcare
Use Case
CT Scan Analysis
Company size
11-50 employees
Headquarters
Houston, Texas, USA
Data labelling
Discover our solutions
About
InformAI is a healthcare informatics company that aims to revolutionize clinical workflows by developing AI solutions that enhance radiologist productivity and accelerate medical diagnosis at the point-of-care. The company is based in the Texas Medical Center, which is the world's largest medical center. Serving 10 million patients and performing 180,000 surgeries annually.
Thanks to the recent funding received from the Cancer Prevention Research Initiative of Texas (CPRIT), InformAI is continually expanding its offerings of AI-enabled solutions in addition to TransplantAI.
The team has developed AI-enabled image classifiers and risk/outcome predictors and is also working on other projects - some of which are still in the pre-market stage. One of their joint projects with V7 is TransplantAI, a solution designed to address the issue of organ non-use caused by sizing mismatches.
The challenge
Size matching of organs between the donor and recipient is a critical issue in the field of transplantation. Proper sizing plays a crucial role in determining the long-term survival of a transplanted organ. One of the most significant challenges in organ transplantation is the high rate of non-use; in some cases the discard rate is can be as high as 40%.
To tackle this challenge, InformAI embarked on a mission to develop a fully automated organ volume estimation model that enables doctors to make accurate side-by-side 3D volumetric comparisons of organs.
To achieve this goal, the team needed a specialized training data platform to annotate CT chest-abdomen-pelvis (CAP) protocol scans and train the model to accurately segment whole organ parenchyma in order to calculate volumes, thus solving the challenge of size mismatches in organ transplantations.
The solution
InformAI leveraged V7's capabilities to develop medical workflows for labeling DICOM files containing organ scans, enabling them to create ground truth for their product, TransplantAI, a comprehensive, integrated AI informatics dashboard that supports clinical workflow for organ transplants.
The user-friendly interface and browser access provided by V7 allowed the team of radiologists to collaborate easily and achieve maximum accuracy more efficiently.
V7 also enabled the team to meet all necessary security and FDA requirements and keep their data safe and confidential by storing it securely in their S3 bucket. Thanks to V7, InformAI was able to bring TransplantAI to life, revolutionizing the field of organ transplantation and providing vital support to healthcare providers and patients alike.
The results
V7 played a crucial role in helping the InformAI team create ground truth, which was used to train a highly accurate model that serves as the foundation for their latest product - TransplantAI.
According to Britton Marlatt, Senior Data Scientist at InformAI:
"Our model achieves 97.0% accuracy, which is a significant improvement considering that transplant clinicians often resort to size matching from demographic information, and this method relies on formulas for estimating organ size with very low coefficients of determination (v2 ~0.35). More advanced methods of size estimation that are presently available heavily rely on a manual process which takes time and valuable resources during the time-critical organ allocation process.This paradigm shift allows us to rapidly calculate the exact size of the organ and perform a side-by-side comparison of the donor and recipient organs."
TransplantAI eliminates the need for approximations and enables medical professionals to accurately determine organ size, making it easier to compare organs between donors and recipients. With the potential to significantly reduce organ non-use rates, TransplantAI is poised to revolutionize the field of organ transplantation.
Training data needs
The global shortage of solid organs for transplantation remains a significant challenge. Demand far outweighs supply by a ratio of 3:1. Although an all time record of over 42,000 transplants were performed in the US in 2022, over 120,000 patients are on the waitlist, with 17 patients dying each day waiting for an organ. Non-use of viable organs is a huge issue, with up to 40% of organs discarded due to substandard matching and sizing methods.
InformAI addressed this issue by developing Transplant AI, a suite of algorithms that includes 3D volumetric comparisons for size matching and better decision-making.
To annotate more than 400 DICOM image series with 170 slices each, the team used V7 and a proprietary dataset from a Texas Medical Center partner consisting of full-body CT scans. Radiology residents labeled heart, lung, kidney, and liver classes, reviewed by a tenured radiologist. InformAI created a streamlined V7 medical workflow for faster QA review cycles.
InformAI then exported the labeled data to train a size estimation model using a U-Net architecture.
Why V7?
InformAI extensively researched training data platforms before choosing V7. The main differentiators were security and accessibility. V7 is differentiated from other labeling platforms by offering secure storage maintained in the client’s AWS S3 bucket, user access controls to enable customized workflows enabled on web-browser as opposed to application software, and 3D medical DICOM image support.
The team required a tool that complies with medical regulations, including FDA & HIPAA, and security standards for hosting sensitive data. V7's compatibility with AWS allowed the team to store partner institution data on S3 servers and view it through V7's API toolkit. V7 also enabled remote onboarding of radiologist labelers, without the need for physical workstations. None of the competitors InformAI reviewed offered this combination.
V7's support for multiple formats, including DICOM, PNG, and NIFTI, was crucial for this annotation project. Easy data import and export options facilitated uploading open-source datasets and exporting labeled data to train models offline.
InformAI’s annotators appreciate V7’s intuitive UI, annotation ease, and flexibility. For maximum accuracy, the team labels all images manually. V7’s interface allows them to use an Apple pencil with an iPad instead of a mouse, contributing to greater convenience and accuracy.