CASE STUDY

Accelerating AI Training through Precise Organ Segmentation​

Client Profile

SliceVault is a cloud-based platform that provides secure and efficient management of sensitive data, particularly in the healthcare industry. SliceVault’s platform is designed to help healthcare organizations improve patient care, reduce costs, and streamline operations by securely storing, managing, and sharing electronic health records (EHRs) and other sensitive information

Challenge

The client needed precise segmentation of different human organs on medical imagery for clinical research and trials. They initially tried to create a small batch with the help of a few researchers, doctors, and radiologists, but it was not financially viable for them to ramp up a team with such specialized skills for the large volume of data needed for the research. The primary goal of the organ segmentation project was to establish a large database of manually segmented organs that can be used to train AI tools. To our knowledge, this is the largest database of high-quality manual annotated data. Professional annotators and routines to check segmentation quality for every organ by dedicated persons contributed to a training set of high quality.

Solution

We started a trial with two annotators who had no medical skills or background to assess how long it would take to train them in annotation based on human anatomy and to understand the medical imagery and differentiate between different organs. After one week of trial, we were confident that we could deliver the tasks with our regular annotators after providing them with some training in human anatomy. After a month, the client chose to scale up the team from two to nine members (seven annotators and two QA). The team was able to minimize the AHT (average handling time) for most organs compared to the client’s set benchmark. Additionally, our QA executives achieved more than 99% accuracy in all organ segmentations to date.
Key Points
  • The client’s involvement early in the project, especially when training and selecting annotators, and also setting up a common benchmark for each type of task.
  • A proper communication process with escalation and screen-sharing sessions with the client for new instructions or edge-case scenarios.
  • Reporting edge-case scenarios to the client and coming up with solutions based on experience and discussions.
  • Creating a compiled version of short instruction manuals and a FAQ document for recording edge-case scenarios for future reference.
  • Detailed trackers and dashboards to monitor progress and identify areas where the team needed assistance.
  • A dedicated QC and QA process to ensure the highest quality of work.

Project Delivery

Orboroi segmented more than 35 different organs in over 12,000 unique cases to provide a comprehensive solution by ensuring incoming images are precisely segmented and thoroughly reviewed for potential errors and are in compliance with medical trial requirements. The skilled annotation team also provided expert suggestions on optimized plans and reports for cost-sufficient and effective results.

Benefits for the client

Organ segmentation is a crucial part of the development of SliceVault's AI tools for the analysis of medical images. This accurate and reliable organ segmentation is being used to train the AI for analyzing medical images and extracting meaningful information that can be utilized to diagnose diseases, develop treatment plans, and monitor patient progress in patient care and in clinical trials. Now the client has a substantial amount of quality datasets for their required research even with a small research budget.

"We receive segmented images of good quality in an efficient way that helps us deliver faster to our clients and grow our business."
- "Lars Edenbrant, Chief Medical Officer, SliceVault"

Conclusion

Manual organ segmentation plays a vital role in the diagnosis and treatment of various medical conditions, and its use has been revolutionized by the advent of machine learning. The use of machine learning algorithms has improved the speed and accuracy of manual organ segmentation, leading to faster and more effective diagnoses and treatments. The impact of manual organ segmentation on medical imaging has been particularly significant in the fields of oncology and neurology, where it has provided insights into the growth and spread of tumors and the progression of neurodegenerative diseases.