Malte Hoffmann

Generalizable AI in neuroimaging @HarvardMed @MGHImaging

[Malte Hoffmann @mh3936]

What's wrong with NIH grants? Science is fundamentally different than remodeling a kitchen. (Mike Lauer) https://t.co/YmwV89vTWL

[Malte Hoffmann @mh3936]

IEEE Signal Processing Magazine published the tutorial on domain randomization for neuroimage generalization https://t.co/XqS2vnXmA7

[Malte Hoffmann @mh3936]

Tired of retraining? This tutorial reviews how domain randomization helps #DeepLearning generalize in neuroimaging: https://t.co/0ZseliaHqw. Accepted to @IEEEspm @IEEEsps | @IEEEBrain @FreeSurferMRI @MGHMartinos @MGBResearchNews

[IEEE Brain @IEEEBrain]

Data? Who needs data? Really, really fascinating presentation by @harvardmed's @mh3936. Note the last bullet on his takeaway slide! Our next speaker is @SandiaLabs' @jbimaknee, who is giving insights into the symbiosis between #NeuromorphicComputing and understanding the #brain.

[IEEE Brain @IEEEBrain]

Here's a comparison of classical versus #DeepLearning for #neuroimaging analysis, courtesy of @mh3936 of @harvardmed. #neurotechnology #neurotech #neuroscience #brain

[IEEE Brain @IEEEBrain]

We're eagerly anticipating our next speaker, @mh3936 of @harvardmed, who will be exploring synthesis-driven #MachineLEarning for neuroimage #analysis. He's thrown out an intriguing statement: you hardly need any data. #neuroimaging #neurotechnology #neurotech #brain

[Malte Hoffmann @mh3936]

New https://t.co/6dt0ll4oBY demo! Improve #MRI-CT #NeuroImage registration by clipping the CT scale to boost soft-tissue contrast @FreeSurferMRI @MGHMartinos @MGH_RI @MIT_CSAIL @AdrianDalca #DeepLearning

[Malte Hoffmann @mh3936]

No retraining! #AI registration tool https://t.co/6dt0ll3QMq lets you tailor warp regularity to your specific data @MGH_RI @MGHMartinos @MIT_CSAIL @AdrianDalca @FreeSurferMRI

[Mass General Brigham Research @MGBResearchNews]

Researchers from @MGHMartinos developed a new approach to image registration, dubbed #SynthMorph, that takes advantage of #AI to achieve registration that is more robust, accurate, and faster than existing methods.

Learn more: https://t.co/RVvT2XE3ZB @mh3936 @MassGeneralNews

[MGH Martinos Center @MGHMartinos]

#SynthMorph, a new, #AI-based approach to image #registration, is more robust, more accurate, and faster than existing methods. Lead author Malte Hoffmann discusses the tool here: https://t.co/nSL02lq3JO

More info: https://t.co/n4WcBxQKeb

@mh3936 @AdrianDalca @FreeSurferMRI

[MIT Jameel Clinic for AI & Health @AIHealthMIT]

Don't move! Blurry MRI scans can lead to misdiagnoses, or repeat scanning. Benjamin Billot, @neeldey, @dc_moyer, @mh3936, Esra Abaci Turk, @BorjanGagoski, Ellen Grant, & #JClinic PI @GollandPolina introduce EquiTrack, to reduce MRI motion blur https://t.co/7SW0spyfbe

[Malte Hoffmann @mh3936]

Explore the new https://t.co/4j4nOioLoF demo on anatomy-aware affine registration! #DeepLearning @MIT_CSAIL @mit_caml @MGHMartinos @FreeSurferMRI @AdrianDalca

[Malte Hoffmann @mh3936]

The 2D and 3D SynthStrip datasets at https://t.co/Zb8xsdhnFN now include label maps for adult MPRAGE scans! @AdrianDalca @MIT_CSAIL @MGHImaging @MGHMartinos @FreeSurferMRI

[Malte Hoffmann @mh3936]

@f_amujo @FreeSurferMRI @AdrianDalca @mit_caml @MIT_CSAIL @MGHMartinos Thanks for your interest in SynthStrip! While the tool was designed for 3D neuroimaging data, it supports stacks of a few coherent slices from 2D scans. For single-slice inputs, you'd likely need to train a custom model.

[Malte Hoffmann @mh3936]

Upgrade your neuroimage analysis! SynthStrip 1.5 now skull-strips multi-frame acquisitions: https://t.co/53DlsPf1ch @FreeSurferMRI @AdrianDalca @mit_caml @MIT_CSAIL @MGHMartinos

[Malte Hoffmann @mh3936]

Passionate about #DeepLearning and #MedicalImaging? Join our team @FreeSurferMRI @MGHMartinos @HarvardMed as a #postdoc, develop cutting-edge algorithms for fetal-brain #MRI. https://t.co/Hnjs55aclz

[Malte Hoffmann @mh3936]

Surprise! SynthStrip can often skullstrip MRI with motion artifacts, despite not seeing this at training. Tool at https://t.co/53DlsPf1ch @FreeSurferMRI @AdrianDalca @mit_caml @MIT_CSAIL @MGHMartinos

[Malte Hoffmann @mh3936]

Awesome talk by @MASI_Lab's @LeonCai at #SPIEMedicalImaging: #DeepLearning image synthesis enables topup distortion correction of fMRI without field map or reverse phase encoding! @FreeSurferMRI @MGHMartinos

[Malte Hoffmann @mh3936]

@soupaulte @FreeSurferMRI @AdrianDalca @MIT_CSAIL @MGHImaging @MGHMartinos It depends. For new anatomy, replacing the label maps might do. We also have an anatomy-agnostic model trained on random shapes; great for fine-tuning! https://t.co/KnCy6Z8isk

[Malte Hoffmann @mh3936]

We made SynthMorph an easy-to-use tool for end-to-end brain registration without preprocessing. Available for #Docker at https://t.co/1I0iXrLtTB and in the nightly @FreeSurferMRI dev build. With @AdrianDalca et al. at @MIT_CSAIL @MGHImaging @MGHMartinos.

[Malte Hoffmann @mh3936]

Make #ML models register specific image parts and adapt to new data types? New #preprint https://t.co/btE2wPUEAD. Code at https://t.co/1I0iXrLtTB. @FreeSurferMRI @MGHMartinos @AdrianDalca @MIT_CSAIL @MIT