Release Name: ADSv1
Notes:
<p>We provide a DL based tool for detection and segmentation of ischemic acute/sub-acute strokes, trained and tested in 2,628 brain MRIs (Liu et al. Nat Commun Med, 2021). Using the original DWI as input, this fully automated system outputs 3D digital infarct mask, volume, and the feature vectors of regions affected by the infarct in two parcellation schemes: structural anatomy and arterial territories. The method is fast (the lesion inference takes 20~30 seconds in CPU; the total processing, including image registration and generation of reports take 3-7 mins, depending on the choice for registration algorithm). ADSv1 includes outputs of the brains and infarct masks mapped to a common space (MNI), ASPECTS calculation, automated radiological reports, with interpretable descriptions of the models' predictions. This system is publicly available, real time, run on local computers, with minimal computational requirements, and is accessible to non-expert users</p>
Changes:
<p>Dec, 04, 2023: We updated this tool to ADSv1.3.zip to fix the following bugs or issues:</p>
<p>1) Hydrocephalus feature vector was fixed.</p>
<p>2) Lesion laterality in radiological report was corrected.</p>
<p>3) Number of Iterations for registration was increased to improve performance </p>
<p>Added an extra example, Subject03, as a not visible (no stroke lesion) case.<br>Thanks for all the past feedback to continue improving this tool. <br>Chin-Fu</p>
<p>--------------------------------------------------------------------<br>Dec,20,2022: We updated this tool to ADSv1.2.zip to fix the following issues in ADSv1.1.zip. <br>1) Computing predict_vol_logml might run into error if no lesion is detected. Now, it's fixed for no lesioned subject. <br>2) The left and right sides flipping issue in array after registration. Now, it's fixed. We appreciate all the feedbacks to continue improving this tool. And we are aiming to refactor and simplify the codes in ADSv2 with our new feature release. <br>Thank you! Chin-Fu<br>----------------------------- <br>We noticed there was a missing template under template folder (AspectsAtlas_padding.nii.gz) in ADSv1.zip. Hence, we released ADSv1.1.zip for updating missing files and new trained models (AA_models_ads). Please download ADSv1.1.zip for future application. If you already downloaded ADSv1.zip, please download ADSv1.1.zip, unzip and replace files under the same ADSv1 folder. Sorry for this inconvenience.</p>
|
