Thesis
Minipig brain MRI: Batten disease phenotypic development and generalized multi-atlas segmentation with joint label fusion
University of Iowa
Master of Science (MS), University of Iowa
Spring 2022
DOI: 10.17077/etd.006428
Abstract
Late-infantile neuronal ceroid lipofuscinosis type 2 (CLN2) is a rare pediatric disease, with symptom development leading to clinical diagnosis. Treatment development requires early diagnosis and effective tracking of progression. Due to the challenging nature of obtaining human CLN2 data, a minipig CLN2 model has been proposed to obtain pre- and postsymptomatic stage data. Minipigs are advantageous for neurological disease insights because they share similar brain structure, tissue composition, and growth to humans. We hypothesize that brain volumetry is valuable in identifying CLN2 early and tracking disease progression in a genetically modified minipig model of CLN2 disease.
Evaluations of CLN2 minipigs and wild type (WT) controls at 12- and 17-months of age were conducted, correlating to early and late stages of disease progression. Acquired data included magnetic resonance imaging (MRI) T1- and T2-weighted data. Segmentations for intracranial brain volume, gray matter, cerebrospinal fluid, white matter, caudate, putamen, and ventricle regions were manually created. Brain regions normalized as a proportion of the intracranial volume were compared between timepoints and cohorts using Gardner-Altman plots, mean differences, and confidence intervals.
At an early stage of disease, the intracranial volume (-9.06cm3), gray matter (-4.29% 95CI: -7.41; -1.83), caudate (-0.15%, 95CI: -0.24; -0.08) and putamen (-0.11%: 95CI: -0.23; - 0.02) were all significantly smaller volumes in CLN2 than WT, while cerebrospinal fluid was larger (+3.42%, 95CI: 2.54; 6.18). As the disease progressed to a late stage, the difference between the gray matter (-8.21%, 95CI: -10.1; -5.56) and cerebrospinal fluid (+6.55%, 95CI: 4.31; 8.51) continued to become more pronounced, while others remained in alignment with the early timepoint.
Tools analogous to human automated segmentation methods need to be created for use in minipig brain volumetry analysis. Manual segmentation labels are the current standard for label volume and region analysis. Creating manual segmentations is labor and time-intensive and subject to error based on intra/inter-rater consistency. A reduction in manual efforts required to create the Batten’s disease segmentations is needed for future minipig neurological studies. A multimodal Joint Label Fusion (JLF) pipeline using both T1- and T2-weighted MRI images and a small number of minipig manual segmentation atlases is proposed to segment minipig MRI images automatically. JLF is a weighted voting method of labeling target images that uses image patch-based comparisons between multiple atlases and a target image to create a single label image for that target.
Testing of the multimodal JLF algorithm used two different pipelines. The first pipeline used single subjects from the CLN2 study as a target, with the remaining CLN2 study subjects used as atlas inputs. The second pipeline used all CLN2 study subjects as atlases, with targets being from a different Huntington’s disease minipig cohort. The resultant JLF label images' accuracy was compared to manual segmentations using dice, balanced Average Hausdorff Distance (bAHD), and volume similarity metrics.
Both the CLN2 and Huntington’s minipig target cohorts created segmentations that were qualitatively similar for all anatomical regions. For the CLN2 target minipig test, the average dice score of all joint label fusion resultant labels was 0.84 (ICV: 0.98, Left Ventricle: 0.89, Right Ventricle: 0.89, Left Caudate: 0.86, Right Caudate: 0.85, Left Putamen: 0.73, Right Putamen: 0.66). For the Huntington’s minipig target pipeline test, the average dice score of all resultant labels was 0.80 (ICV: 0.95, Left Caudate: 0.80, Right Caudate: 0.80, Left Putamen: 0.58, Right Putamen: 0.38). Additional bAHD and volume similarity scores metrics further supported the dice correlation scores for all both the CLN2 and Huntington’s minipig cohorts.
MRI brain volumetry in this minipig model of CLN2 is sensitive to early disease detection and longitudinal change monitoring, providing a valuable tool for pre-clinical treatment development and evaluation. The combination of high-quality T1- and T2-weighted CLN2 minipig images and the application of the JLF minipig pipeline showed promising results in creating automated segmentations for both target minipig cohorts of different diseases and species. JLF demonstrated utility for speeding up segmentation for volumetric analysis for future neurological studies. The JLF speedup will help with disease phenotyping, treatment development, and pre-clinical intervention and application to humans.
Details
- Title: Subtitle
- Minipig brain MRI: Batten disease phenotypic development and generalized multi-atlas segmentation with joint label fusion
- Creators
- Kevin Knoernschild
- Contributors
- Jessica Sieren (Advisor)Hans J Johnson (Advisor)Joseph Reinhardt (Committee Member)Sajan Lingala (Committee Member)Michelle Voss (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Biomedical Engineering
- Date degree season
- Spring 2022
- DOI
- 10.17077/etd.006428
- Publisher
- University of Iowa
- Number of pages
- xviii, 131 pages
- Copyright
- Copyright 2022 Kevin Knoernschild
- Language
- English
- Description illustrations
- Illustrations, charts, graphs, tables
- Description bibliographic
- Includes bibliographical references (pages 128-131).
- Public Abstract (ETD)
Neurological diseases are a common occurrence in humans, however not all neurological diseases are easily studied. Battens disease is a rare neurological disease that impacts children and for which systematic magnetic resonance imaging (MRI) data is limited. Due to the difficulty of obtaining data for Battens disease, a minipig disease study was proposed. Minipig brains are like human brains in structure and development, meaning that changes over time can be studied and findings can be transferred back to humans in clinical application.
This minipig study used systematic MRI to measure targeted portions of the brain in the Battens diseased minipigs and with comparisons to healthy minipigs. The results generated showed that at an early stage of disease development the Battens disease minipigs had significant differences in regions of brain and that some of these differences became more pronounced as the disease progressed over time (12 to 17 months of age). Additionally, a process was implemented to automate labeling of minipig brain regions in the future. For the Battens disease study, a human drew the minipig label outlines by hand so that volumes could be calculated for all subjects. This hand drawn labeling was time consuming and limited the ability to do timely analysis of MRIs. An automated labeling process was developed using the Batten disease images and hand drawn labels to label new minipig brain MRIs automatically. This study demonstrated indicators of early Battens disease and its progression using MRI, combined with the automated brain labelling process developed, the minipig model will be valuable in the development and testing of treatments for humans.
- Academic Unit
- Roy J. Carver Department of Biomedical Engineering
- Record Identifier
- 9984271452002771
Metrics
55 File views/ downloads
64 Record Views