Any artificial intelligence algorithm needs to "learn" from established examples, which are called "ground truth." Our contention is that, the higher the quality of the ground truth used to train an algorithm, the better the performance of that algorithm, all else being equal.
In order to test the first step in this logic, we performed an objective study with 11 neuro-radiologists and radiation oncologists participating in rating a randomized 20-study sample of segmentations of 100 different brain tumor studies. Each study compared Neosoma ground truth to ground truth from the Brain Tumor Segmentation Challenge.
The Results Are In
Neosoma wins out the majority of the time
The results speak for themselves, as you can see in the graph below: In 70.1% of the cases, Neosoma's ground truth was preferred outright over the BraTs segmentation, with another 9.5% undecided.
These results are important to validate our efforts in developing high-quality ground truth data sets to train our solutions, which in turn leads to high-quality, clinically-beneficial products.
