Computer Vision News - October 2016

Challenge Magnetic Resonance Imaging (MRI) is the reference standard used to assess those measures, also due to the reproducibility of making these important measurements. The challenge is to help replace the slow manual process used by the doctor to derive the ejection fraction with an automatized procedure, which would arrive to the same conclusion in a more efficient way. Participant were requested to create an algorithm which would automatically measure end-systolic and end-diastolic volumes in MRIs, examining cardiac images from more than 1,000 patients. Results A relatively short development time was allotted for the challenge, nonetheless the top teams obtained excellent results. This is particularly notable for those who hand-labeled patient data without being trained physicians. You can read here further medical perspectives on the results of this challenge by Andrew Arai, MD (NIH, Bethesda). Coming to the leading teams submission analysis , an EF under 35% is a dire emergency, around 60% is normal, and above 73% is considered hyperactive. The good news is that the leading models reviewed keep the diagnosis categories quite tightly grouped together: even though they are not always perfectly precise, there is a very low probability that an emergency EF will be incorrectly categorized in the mild or the average categories. The normal to mild diagnoses are very likely to stay within their domain of the matrix. Michael Hansen, co-PI (Principal Investigator) for this Data Science Bowl, noted that the best models can fail, but they should “ fail loudly ”: in other words, they should be able to predict their own level of accuracy, so that when the system flags the confidence in the prediction as insufficient, it is possible to consider retaking a measurement. Results show low correlations between the model error and “confidence” distributions, suggesting that further improvement could be done in assessing the prediction value. “The leading models reviewed keep the diagnosis categories quite tightly grouped together” Challenge Computer Vision News Challenge 27 Data science applied to cardiology can help physicians save more lives