DS6, Deformation-aware Semi-supervised Learning: Application to Small Vessel Segmentation with Noisy Training Data

Soumick Chatterjee, Kartik Prabhu, Mahantesh Pattadkal, Gerda Bortsova, Chompunuch Sarasaen, Florian Dubost, Hendrik Mattern, Marleen de Bruijne, Oliver Speck, Andreas Nürnberger

Keywords: UNet Multi-scalce Supervision, 7 Tesla MRA, TOF-MRA, Imperfect ground-truth

TL;DR: UNet multi-scale supervision with deformation consistency in a semi-supervised manner improves small vessel segmentation from 7T TOF-MRA

Abstract: The advancement of 7 Tesla MRI systems enabled the depiction of very small vessels in the brain. Segmentation and quantification of the small vessels in the brain is a critical step in the study of Cerebral Small Vessel Disease, which is a challenging task. This paper proposes a deep learning based on U-Net Multi-Scale Supervision architecture to automatically segment small vessels in 7 Tesla 3D Time-of-Flight (TOF) Magnetic Resonance Angiography (MRA) data trained on a small imperfect semi-automatically segmented dataset and was made equivariant to elastic deformations in a self-supervised manner using deformation-aware learning to improve the generalisation performance. The proposed method achieved a dice score of 80.44$\pm$0.83 while being compared against the semi-automatically created labels and 62.07 while comparing against manually segmented region.

Paper Type: both

Primary Subject Area: Segmentation

Secondary Subject Area: Learning with Noisy Labels and Limited Data

Paper Status: based on accepted/submitted journal paper

Source Code Url: https://github.com/soumickmj/DS6

Data Set Url: Data from the paper "Prospective motion correction enables highest resolution time‐of‐flight angiography at 7T" by Mattern et al. 2018 (Magnetic resonance in medicine) was used in the research. Data can not be made available due to subject privacy reasons.

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.