Detecting cancer with superhuman accuracy
Building a computer vision system that detects lung cancer nodules with superhuman accuracy — leading to FDA 510(k) clearance.
83%
Sensitivity in detecting lung nodules
30%+
More sensitive than radiologists alone
11,000+
Patient images in validation study
FDA 510(k)
Clearance achieved
About Imidex
IMIDEX is an Artificial Intelligence Diagnostics solution provider for lung cancer detection. The company has developed a software-as-a-medical-device solution that enables health providers to better detect nodules that can lead to lung cancer discovery.
According to the American Cancer Society, lung cancer is the leading cause of cancer death, making up almost 25% of all cancer deaths. Each year, more people die of lung cancer than of colon, breast, and prostate cancers combined. IMIDEX was founded on the mission to save lives using data science, with the goal of becoming the standard of care for lung cancer.
Situation
To meet U.S. Food and Drug Administration (FDA) approvals to bring IMIDEX's VisiRad solution to market, the company was required to showcase the sensitivity modeling of its lung cancer nodule detection. The goal was to increase identification of lung cancer nodules well beyond the average of a radiologist's sensitivity.
IMIDEX had collected thousands of chest x-rays, which were reviewed manually by a clinical team. It needed a data science partner to build a machine learning solution that would accelerate the review of x-rays without compromising the sensitivity of nodule detection.
Solution
Echelon's team developed a convolutional neural network that provided a superior way of identifying nodules that well exceeded the sensitivity goals. The solution enabled IMIDEX to reach its solution review deadline and allowed for an advancement to testing specificity goals.
The Echelon solution is now integrated into the existing workflow to ingest x-rays before being passed to IMIDEX's Clinical Integrity Team. As more images are added to the growing database, the machine learning model continues to increase the diagnostic accuracy to detect lung nodules.
With minimal input, the system automatically adjusts hyperparameters, performs processes en masse through pipeline components, and validates results. This allows teams with excellence in data science to focus their efforts on higher-value tasks, such as identifying and correcting anomalies in the data and pinpointing subsets of data that need more attention.
Results
IMIDEX achieved FDA 510(k) clearance for VisiRad XR, making it one of the first FDA-cleared medical devices built on Google Vertex AI. In a retrospective standalone study utilizing over 11,000 patient images, VisiRad XR detected lung nodules and masses at a sensitivity of 83% with a fixed false positives per image rate.
Clinical evaluation was performed through a multi-reader, multi-case clinical validation study using 600 images and 24 radiologists from across the country. VisiRad XR improved each reader's ability to detect pulmonary nodules and masses, demonstrating statistically significant improvement in AUC and increased sensitivity across all readers, regardless of experience level, specialty, or training background.
For an average hospital in the US performing 50,000 chest X-rays annually, VisiRad XR could identify up to an additional 750 lung nodules or masses per year.
Further Reading
Have a similar challenge?
Let’s Build Something Together
Tell us about your problem and we’ll set up a time to talk.