Humans teaching machines to help humans

Purpose-driven technology for unlimited opportunity

Case Study – Improving Medical Imaging AI with HEWMEN

Using HEWMEN to Enhance DNN Image Segmentation of Age Related Macular Degeneration OCT images.

Age-related Macular Degeneration (AMD) is a disease of the retina for which there is no cure. It is the leading cause of vision impairment in those aged 50 and older. Identifying those at risk, early diagnosis, and tracking treatment effectiveness is crucial.

Optical Coherence Tomography (OCT) is an imaging method used to diagnose and track treatment of AMD. Automated tools to analyze OCT retinal scans exist, but human experts must modify the images before use. This is time-consuming, expensive, and inefficient.

BALANCED deployed its HEWMEN platform behind an online video game built around OCT retinal scans to facilitate human guided AI to train a deep learning algorithm used to diagnose and track the treatment effectiveness of AMD.

This refined algorithm significantly improves diagnosis and management of AMD, saving time, and expense. Because HEWMEN is data and algorithm agnostic, it can be used in many fields allowing for greater algorithm efficiency without requiring experts.

Partners

BALANCED Solution - HEWMEN Platform Benefits

Improving Model Accuracy

Increased algorithm precision (+78%) and accuracy (+38%)

Reducing Model Size

Decreased computational requirements with increased model accuracy and precision

Leveraging Small Data Sets

Improved algorithm performance vs traditional training models requiring large data sets

Refocusing Talent

Improved algorithm automation allowing experts to pursue more complex problems

Improving Margins

Decreased specialists referrals, faster training time, and reduced development and operating costs

BALANCED'S Mission

To get better treatments to patients faster by enabiling AI

Methodology

Automated disease detection via Human in the loop (HiTL) Machine Learning and online gaming

OCT Raw Data OCT Data Gamified Slide to compare Slide to compare

Results from HEWMEN AMD Case Study

Players through human computational gaming (HCG) effectively trained an existing deep learning algorithm to identify retinal disease

Conclusion

"With interaction between human intuition and machine learning, the way we perform clinical trials in 3-5 years will be very different from the way it is done today."

- Karl G. Csaky, MD, PhD, Managing & Medical Director, Retina Foundation of the Southwest

Through the HEWMEN platform, BALANCED has a unique approach to training AI algorithms. Our ability to do so using smaller data sets with greater accuracy and precision at higher margins than conventional techniques provides an exclusive advantage to our customers/partners.

robortAnimConclusion

Our Published Papers

  1. C. Clark and M. Ouellette, “Using Human Computation Game-Based Input to Enhance DNN Image Segmentation of Age-Related Macular Degeneration OCT Images with Small Datasets,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020 (Under Review)

  2. C. Clark, M. Ouellette, and K. Csaky, “Training Players to Analyze Age-Related Macular Degeneration Optical Coherence Tomography Scans using a Human Computation Game,” in 2019 IEEE 7th International Conference on Serious Games and Applications for Health (SeGAH), Kyoto, Japan, Aug. 2019, pp. 1–7, doi: 10.1109/SeGAH.2019.8882430.

  3. C. Clark, I. Greenberg, and M. Ouellette, “A model for integrating human computing into commercial video games,” in 2018 IEEE 6th International Conference on Serious Games and Applications for Health (SeGAH), Vienna, May 2018, pp. 1–8, doi: 10.1109/SeGAH.2018.8401316.

  4. C. Clark and M. Ouellette, “Video games as a distributed computing resource,” in Proceedings of the International Conference on the Foundations of Digital Games - FDG ’17, Hyannis, Massachusetts, 2017, pp. 1–7, doi: 10.1145/3102071.3102099.