A Dempster-Shafer approach to trustworthy AI with application to fetal brain MRI segmentation

Fidon, Lucas; Aertsen, Michaël; Wiestler, Benedikt; Bink, Andrea; David, Anna L.; Deprest, Thomas; Emam, Doaa; Guffens, Fr/'ed/'eric; Jakab, András; Kasprian, Gregor; Kienast, Patric; Melbourne, Andrew; Menze, Bjoern; Mufti, Nada; Pogledić, Ivana; Prayer, Daniela; Marlene, Stuempflen,; Elslander, Esther Van; Ourselin, Sébastien; Deprest, Jan; Vercauteren, Tom

doi:10.48550/arxiv.2204.02779

Cited by 3 publications

(4 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…DRO with a ϕ-divergence reweights the examples in the training dataset but cannot account for subsets of the true distribution that are not represented at all in the training dataset. We investigate this problem in our following work (Fidon et al, 2022).…”

Section: Discussionmentioning

confidence: 99%

Distributionally Robust Deep Learning using Hardness Weighted Sampling

Fidon¹,

Ourselin²,

Vercauteren³

2022

Melba

View full text Add to dashboard Cite

Limiting failures of machine learning systems is of paramount importance for safety-critical applications. In order to improve the robustness of machine learning systems, Distributionally Robust Optimization (DRO) has been proposed as a generalization of Empirical Risk Minimization (ERM). However, its use in deep learning has been severely restricted due to the relative inefficiency of the optimizers available for DRO in comparison to the wide-spread variants of Stochastic Gradient Descent (SGD) optimizers for ERM.<br>We propose SGD with hardness weighted sampling, a principled and efficient optimization method for DRO in machine learning that is particularly suited in the context of deep learning. Similar to a hard example mining strategy in practice, the proposed algorithm is straightforward to implement and computationally as efficient as SGD-based optimizers used for deep learning, requiring minimal overhead computation. In contrast to typical ad hoc hard mining approaches, we prove the convergence of our DRO algorithm for over-parameterized deep learning networks with ReLU activation and finite number of layers and parameters.<br>Our experiments on fetal brain 3D MRI segmentation and brain tumor segmentation in MRI demonstrate the feasibility and the usefulness of our approach. Using our hardness weighted sampling for training a state-of-the-art deep learning pipeline leads to improved robustness to anatomical variabilities in automatic fetal brain 3D MRI segmentation using deep learning and to improved robustness to the image protocol variations in brain tumor segmentation.a decrease of 2% of the interquartile range of the Dice scores for the enhanced tumor and the tumor core regions.<br>Our code is available at <a href='https://github.com/LucasFidon/HardnessWeightedSampler'>https://github.com/LucasFidon/HardnessWeightedSampler</a>

show abstract

Section: Discussionmentioning

confidence: 99%

Distributionally Robust Deep Learning using Hardness Weighted Sampling

Fidon¹,

Ourselin²,

Vercauteren³

2022

Melba

View full text Add to dashboard Cite

show abstract

“…Automatic segmentation of the unmyelinated white matter, ventricles and cerebellum was computed using deep‐learning‐based methods 40 , 41 , 42 . This provided good‐quality initial segmentation, which was subsequently corrected manually by a trained obstetrician (N.M.) and corrected as necessary thereafter by a consultant board‐certified pediatric neuroradiologist (M.A.)…”

Section: Methodsmentioning

confidence: 99%

Assessment of longitudinal brain development using super‐resolution magnetic resonance imaging following fetal surgery for open spina bifida

Mufti

Chappell

Aertsen

et al. 2023

Ultrasound in Obstet & Gyne

View full text Add to dashboard Cite

ObjectivesFetal surgery is offered in selected fetuses with open spina bifida (OSB) to improve long‐term outcome. Using advanced Magnetic Resonance Imaging (MRI) quantification of volume, surface area, shape of cerebral structures, and surface curvature analysis using parameters that correspond to gyrification, we studied the effect of fetal surgery on brain development.MethodsWe compared MRI images of 29 OSB fetuses before fetal surgery (<24 weeks), one‐ and six‐weeks post‐surgery, to 36 age‐matched normal controls (21+2‐36+2 weeks). Automated super‐resolution reconstruction provided 3D isotropic volumetric brain images. Unmyelinated white matter, cerebellum and ventricles were automatically segmented and manually refined; cerebral volume, surface area, and shape parameter (volume/surface area) were quantified. Mathematical markers (shape index (SI) and curvedness) were used to measure gyrification. Parameters were assessed according to lesion type (e.g., myeloschisis (MS)), persistent postoperative hindbrain herniation (HH), and presence of supratentorial anomalies namely partial agenesis of the corpus callosum (pACC), and heterotopia (HT).ResultsOSB ventricular volume growth (mm3/week) and shape parameter (mm/week) were higher six weeks post‐surgery compared to controls; (2500.94, IQR: 1665.51‐3604.1 mm3/week vs 708.21, IQR: 435.06‐965.77 mm3/week, p<0.001), and (0.075, IQR: 0.047‐0.112 mm/week vs 0.022, IQR: 0.009‐0.042 mm/week, p=0.046). Ventricular volume growth increased six weeks post‐surgery in cases with pACC (p<0.001) and persistent HH (p=0.002). During that time period, OSB unmyelinated white matter shape parameter change (mm/week) decreased compared to controls (0.056, IQR: 0.044‐0.092 mm/week vs 0.159, IQR: 0.1‐0.247 mm/week, p=0.002) particularly in cases of persistent HH (p=0.011), MS (p=0.015), HT (p=0.022), HT with CC anomalies (p=0.017), and persistent HH with CC anomalies (p=0.007). Six weeks postoperatively despite a remarkable reduction in curvedness/week (mm‐1) (0.061, IQR: 0.040‐0.093 mm‐1/week vs 0.094, IQR: 0.072‐0.145 mm‐1/week, p<0.001) compared to the initial increase one week after (0.144, IQR: 0.101‐0.233 mm‐1/week vs 0.072, IQR: 0.059‐0.08 mm‐1/week, p<0.001), gyrification using SI appeared increased overall compared to controls. This observation was more prominent in fetuses with pACC, and severe ventriculomegaly (p‐value range=0.006 to<0.001).ConclusionsFollowing fetal OSB repair, volume, shape and curvedness in ventricles and unmyelinated white matter were significantly different from normal fetuses. Morphological brain changes after fetal surgery were not limited to effects on CSF circulation. These observations may have neurocognitive implications for postnatal outcomes.This article is protected by copyright. All rights reserved.

show abstract

“…The 40 3D MRIs and original segmentations (as provided with the FeTA dataset) were inspected by two paediatric radiologists within our institutions, MA and PD, with more than 8 years of experience in segmenting fetal brains. Corrections of the segmentations were performed 26 – 28 to reduce the variability against the published segmentation guidelines that was released with the FeTA dataset 23 , 25 . Two volumes of spina bifida aperta cases ( and ) were excluded because the poor quality of the 3D reconstruction did not allow to segment them reliably for the seven tissue types.…”

Section: Methodsmentioning

confidence: 99%

“…The manual segmentations for the fetal brain MRI of FeTA dataset, that we have contributed in our previous work 26 – 28 , are publicly available on Zenodo: https://doi.org/10.5281/zenodo.6878474 51 under the term of the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported license (CC BY-NC-ND 3.0). Access to the data is restricted.…”

Section: Data Availabilitymentioning

confidence: 99%

A spatio-temporal atlas of the developing fetal brain with spina bifida aperta

Fidon¹,

Viola²,

Mufti³

et al. 2022

Open Res Europe

Self Cite

View full text Add to dashboard Cite

Background: Spina bifida aperta (SBA) is a birth defect associated with severe anatomical changes in the developing fetal brain. Brain magnetic resonance imaging (MRI) atlases are popular tools for studying neuropathology in the brain anatomy, but previous fetal brain MRI atlases have focused on the normal fetal brain. We aimed to develop a spatio-temporal fetal brain MRI atlas for SBA. Methods: We developed a semi-automatic computational method to compute the first spatio-temporal fetal brain MRI atlas for SBA. We used 90 MRIs of fetuses with SBA with gestational ages ranging from 21 to 35 weeks. Isotropic and motion-free 3D reconstructed MRIs were obtained for all the examinations. We propose a protocol for the annotation of anatomical landmarks in brain 3D MRI of fetuses with SBA with the aim of making spatial alignment of abnormal fetal brain MRIs more robust. In addition, we propose a weighted generalized Procrustes method based on the anatomical landmarks for the initialization of the atlas. The proposed weighted generalized Procrustes can handle temporal regularization and missing annotations. After initialization, the atlas is refined iteratively using non-linear image registration based on the image intensity and the anatomical land-marks. A semi-automatic method is used to obtain a parcellation of our fetal brain atlas into eight tissue types: white matter, ventricular system, cerebellum, extra-axial cerebrospinal fluid, cortical gray matter, deep gray matter, brainstem, and corpus callosum. Results: An intra-rater variability analysis suggests that the seven anatomical land-marks are sufficiently reliable. We find that the proposed atlas outperforms a normal fetal brain atlas for the automatic segmentation of brain 3D MRI of fetuses with SBA. Conclusions: We make publicly available a spatio-temporal fetal brain MRI atlas for SBA, available here: https://doi.org/10.7303/syn25887675. This atlas can support future research on automatic segmentation methods for brain 3D MRI of fetuses with SBA.

show abstract

A Dempster-Shafer approach to trustworthy AI with application to fetal brain MRI segmentation

Cited by 3 publications

References 45 publications

Distributionally Robust Deep Learning using Hardness Weighted Sampling

Distributionally Robust Deep Learning using Hardness Weighted Sampling

Assessment of longitudinal brain development using super‐resolution magnetic resonance imaging following fetal surgery for open spina bifida

A spatio-temporal atlas of the developing fetal brain with spina bifida aperta

Contact Info

Product

Resources

About