AI-Enabled Digital Pathology
for Cancer Diagnosis
GREATER
Laboratory Efficiency
This public health institution, responsible for biomedical and public health research, depends on iMerit to train pathology algorithms for improved cancer treatment.
Challenge
During treatment, cancer patients routinely undergo biopsies to measure disease progression and treatment efficacy. When biopsies are sent to a lab, pathologists manually examine cells through microscopes.
It can often take up to 60 days for a sample to be turned around, resulting in poorer treatment responsiveness and manpower optimization. Patients suffered due to delayed results, which gave their disease more time to progress with less time to alter treatment.
After examining digital pathology solutions, this major public health institution chose to build an Al model that could automatically scan, annotate, and classify cells to boost lab efficiency and patient care. When in-house pathologists tried annotating slide images, the sheer size of the files caused their annotation software to crash.
Realizing they needed a creative solution, this institution began evaluating annotation vendors.
“We were still doing things the old fashioned way, and needed a tech-forward solution to improve our turnaround times. ”
Solution
To annotate the images, this institution chose iMerit for its HIPAA-compliant processes, US-based team, and diverse team of cancer and pathology experts. To solve the file size challenge, iMerit suggested dividing images into quadrants and sending them over one at a time.
To optimize annotation costs, iMerit recommended specialized medical annotators instead of expensive medical experts. Working with the client, iMerit developed a specialized medical curriculum that taught annotators to accurately identify and annotate cancer cell nuclei.
After performing the annotations, ground-truth datasets were generated to begin training the model.
“It was a night and day difference. Our labs were turning samples around faster than ever. ”
Result
Within two months, the model was conducting annotations with 97% accuracy at a 10x faster pace compared to pathologists working in a lab. The accelerated sample turnaround times meant faster disease staging and treatment evaluation for patients.
As the model performed annotations without the need of a doctor-in-the-loop, the client could allocate their pathology teams to other projects. Today, this institution is ramping up its partnership with iMerit across other pathology and radiology projects.
