Multiple Instance Learning and Local Attention for Ischemic Stroke Infarct Segmentation on Early Acute Baseline CTA
Keywords: CT Angiography, Stroke, Infarct, Multiple Instance Learning, Attention
Abstract: CT is the most available imaging modality for assessing the condition of acute ischemic stroke patients. However, early signs of infarction are barely visible on CT scans, and expert infarct assessments on these images are highly variable. We investigate whether weak labels in conjunction with Multiple-Instance Learning (MIL) can be used to segment early signs of infarction on baseline CTA scans. We propose an alternative to the standard MIL attention, local attention, which exploits the symmetry of the brain. We compare our infarct volume predictions with $3$ clinically validated CTP software packages and observe that unlike standard attention, local attention’s performance is comparable with the CTP software. This MIL approach including local attention therefore allows for infarct assessment on CTA without implementing the burdensome and time-demanding CTP scans.
Paper Type: both
Primary Subject Area: Segmentation
Secondary Subject Area: Application: Radiology
Paper Status: original work, not submitted yet
Source Code Url: The research was done with company resources of Nico.Lab and we do not have their permission to publish the source code at this point.
Data Set Url: We do not have permission to publish the dataset.
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
TL;DR: Using weakly supervised learning and an alternative to MIL attention to segment the ischemic stroke infarct on early acute baseline CTA scans.
4 Replies
Loading