2/7 EchoNet-Dynamic can predict EF with human accuracy (mean error of 4.1%), accurately identify HFrEF (AUC = 0.97), and is more consistent than humans on prospective repeat testing.pic.twitter.com/WyfANFuFNT
Prikaži ovu nit
1/7 We’d like to share something interesting for machine learning in cardiology. We trained a neural network, EchoNet-Dynamic, to interpret #echofirst videos to trace the left ventricle and estimate ejection fraction. @bryandhe @euanashley @james_y_zou The results are striking:pic.twitter.com/DKWZ8A92Gf
-
-
-
3/7 Using past human labels from our echo reporting database, we precisely perform semantic segmentation of the left ventricle. DSC = 0.92. This beat-by-beat, every frame, frame level labeling would be too laborious for humans in clinical practice!
Prikaži ovu nit -
4/7 With this precision phenotyping, we can quantify beat-to-beat variation across the video and see more variation in AF and ectopy.pic.twitter.com/BlzNUgOwEj
Prikaži ovu nit -
5/7 EchoNet-Dynamic was implemented in
@pytorch and the code is available at:https://github.com/douyang/EchoNetDynamic …Prikaži ovu nit -
6/7 And thanks to our friends
@StanfordAIMI@curtlanglotz@mattlungrenMD, we will be releasing our dataset before the#ML4H workshop at@NeurIPSConf. 10k deidentified A4c echo videos used to train EchoNet-Dynamic as a resource for ML.Prikaži ovu nit -
7/7 Tremendous thanks to
@HeartBobH ,@GhorbaniAmirata and the@StanfordMed#echofirst lab.@paheidenreich David Liang, Ingela Schnittger.@ASE360@StanfordHealth@StanCVFellows@StanfordDeptMedPrikaži ovu nit -
8/7 (Oops!) Our
@medrxivpreprint is athttps://www.medrxiv.org/content/10.1101/19012419v2 …Prikaži ovu nit
Kraj razgovora
Novi razgovor -
-
-
This is excellent work! Congratulations!! I’ve never seen Echo software auto-quant the beat-to-beat variation in AF the way you did with
#EchoNet-Dynamic. Very impressive. In NSR, how does your software compare to, say,@PhilipsEcho HeartModel software that comes on the Epiq? -
I don’t have much experience with HeartModel but would love to compare!
@PhilipsEcho
Kraj razgovora
Novi razgovor -
-
-
Congrats on this impressive work!
-
Thanks Wunan!
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.