Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Investigation of Image Co-Registration for Mapping Cortical Microstructures using dMRI

Alm Ekwall, Julia (2024) MSFT02 20241
Medical Physics Programme
Abstract
Introduction and aim
The changes in the brain caused by Alzheimer’s disease usually begin many years before the first symptoms arise. The effect of the neurodegenerative disease involves both morphological and microstructural changes in the brain, due to neurons dying. These changes in the microstructure can be measured with diffusion MRI. The resolution however limits the accuracy of the imaging, making structures more difficult to distinguish due to partial volume effects (PVE). PVE occurs when voxels contain information from multiple tissues, which takes place more frequently when the resolution is low. To reduce the PVE the diffusion-weighted image (DWI) can be registered to a T1-weighted image. The registration enables the use of a... (More)
Introduction and aim
The changes in the brain caused by Alzheimer’s disease usually begin many years before the first symptoms arise. The effect of the neurodegenerative disease involves both morphological and microstructural changes in the brain, due to neurons dying. These changes in the microstructure can be measured with diffusion MRI. The resolution however limits the accuracy of the imaging, making structures more difficult to distinguish due to partial volume effects (PVE). PVE occurs when voxels contain information from multiple tissues, which takes place more frequently when the resolution is low. To reduce the PVE the diffusion-weighted image (DWI) can be registered to a T1-weighted image. The registration enables the use of a segmentation of the gray and white matter from the T1-weighted image. This segmentation can be used on the DWI to measure changes in the microstructure of the brain. There are however many ways to register images, therefore there is a need to identify which method of registration that produces the best alignment between images with the least influence of PVE. There are additional steps on how to reduce the PVE, such as using a Bowsher prior method where the registered diffusion images are reconstructed into a higher resolution.

Material and Methods
Registrations were performed in different ways by using the different transforms rigid, affine and nonlinear. A combined method with multiple input images was also used to perform a registration in this work. The registrations were alternately driven by a b0-image or a b2500-image. The b0 and b2500 are images with different contrasts, where the b0 has a low diffusion weighting and the b2500 has a high diffusion weighting. The registrations were evaluated by visual evaluations and by extracting mean diffusivity (MD) values from the cortex. The MD-values in the cortex range up to approximately 1.1 · 10−3 mm2/s for a cognitively normal individual. Values higher than this occur due to PVE with the cerebrospinal fluid (CSF) surrounding the cortex. The fraction of voxels in the cortex with an MD-value above 1.1 · 10−3 mm2/s was therefore used as a PVE-index. Furthermore, the distribution of PVE-indices was evaluated in 70 cortical regions which were extracted with the Desikan–Killiany atlas available in FreeSurfer. A reconstruction of the registered diffusion-weighted images was also performed with the Bowsher prior method.

Results
The registrations showed a notable difference in the visual evaluations, especially when the b0 was driving the registrations. From the evaluation of the PVE-index, the results indicated that it was superior to drive the registration with the b2500 instead of the b0. A paired t-test confirmed that there was a statistically significant difference between driving the registrations using either the b0 or the b2500. The PVE-indices varied between different registrations as well. The difference was more notable between registrations when the b0 was used in comparison to the b2500. A two-way ANOVA without reproducibility showed a statistically significant difference between registrations for both b0 and b2500 driving the registrations. The ANOVA was followed up with Tukey’s post hoc honestly significant difference (HSD) test. It showed a statistically significant difference between all registrations when the b0 was used. For when the b2500 was driving the registrations, all registrations showed a statistically significant difference except for between the rigid and affine and between the affine and nonlinear transforms. The reconstructed diffusion images with the Bowsher prior method resulted in a visually higher resolution. The algorithm however altered the values in the images which prevented further evaluation of this approach.

Conclusion
The results indicated that the rigid transform produced the highest registration accuracy when the b0 was used to drive the registrations. When the b2500 was used to drive the registrations, the accuracy of the registrations increased and the result suggested that the accuracy was similar for rigid, affine and nonlinear. The result also indicated that the analysis of registration accuracy is not straightforward. (Less)
Popular Abstract (Swedish)
Alzheimers sjukdom är idag den vanligaste neurodegenerativa sjukdomen i världen. Det klassas som en ålderssjukdom, och med en ökande medelålder i befolkningen, ökar således risken för fler att drabbas. Sjuk- domen orsakar en förtunning av hjärnvävnad och presenterar sig med symptom så som minnesstörningar som gradvis blir värre. Symptomen av Alzheimers sjukdom kan däremot uppkomma så sent som flera år efter att sjukdomen har börjat bryta ned hjärnan. På grund av det så finns det ett behov och motivation för att hitta indikationer av Alzheimers sjukdom i en befolkning under ett tidigare skede. Detta för att kunna påbörja behandling tidigare men även för att möjliggöra en övervakning av sjukdomsförloppet för att fördjupa kunskaperna om... (More)
Alzheimers sjukdom är idag den vanligaste neurodegenerativa sjukdomen i världen. Det klassas som en ålderssjukdom, och med en ökande medelålder i befolkningen, ökar således risken för fler att drabbas. Sjuk- domen orsakar en förtunning av hjärnvävnad och presenterar sig med symptom så som minnesstörningar som gradvis blir värre. Symptomen av Alzheimers sjukdom kan däremot uppkomma så sent som flera år efter att sjukdomen har börjat bryta ned hjärnan. På grund av det så finns det ett behov och motivation för att hitta indikationer av Alzheimers sjukdom i en befolkning under ett tidigare skede. Detta för att kunna påbörja behandling tidigare men även för att möjliggöra en övervakning av sjukdomsförloppet för att fördjupa kunskaperna om sjukdomen. Det som faktiskt sker i hjärnan är att neuroner dör, vilket orsakar en mikrostrukturell förändring i de drabbade områdena. Orsaken till varför detta sker är dock ännu inte fastställd. Denna förändringen i hjärnan visar sig dock vara mätbar med magnetresonanstomografi (MRT) eller mer specifikt diffusions MRT.

Det finns däremot begränsningar av upplösningen i bilder som producerats med diffusions MRT. Upplös- ningen är för låg för att kunna urskilja mindre strukturer i bilden och istället blir det man ser påverkat av så kallade partiella volymseffekter (PVE). PVE innebär att strukturerna även innehåller information från omkringliggande strukturer, för att upplösningen inte är hög nog för att urskilja dem. Då ändringarna i hjärnan orsakade av Alzheimers sjukdom sker på en mikroskopisk nivå så behöver mätmetoderna vara mer exakta. En lösning på det problemet är att man använder sig av bildregistrering. En bildregistrering innebär att man för över en diffusionsbild in i rymden av en högre upplöst bild, exempelvis en T1-viktad MRT-bild. Om bilderna befinner sig i samma rymd kan man dra nytta av den högre upplösta bilden för att öka up- plösningen av diffusionsbilden. Detta kommer då resultera i att de partiella volymseffekterna upplevs som mindre. Bildregistrering är däremot en metod i sig som kräver optimering. Det finns många olika sätt en bildregistrering kan utföras på men vilket sätt som är det bästa är ännu okänt. Utvärderingsverktyg som evaluerar prestandan i bildregistreringen finns det också ett behov av. Således har syftet med detta arbetet varit att jämföra olika metoder för registrering för att eventuellt komma fram till en metod som ger oss en bildregistrering med bästa möjliga resultat. Förrutom det så har syftet även varit att hitta ett sätt att utvärdera registreringarna på.

De olika metoderna för att utföra en registrering på som har utvärderats i detta arbetet är: rigid, affine, nonlinear (ickelinjär) och combined som är en kombinerad metod (innehållande en rigid, affine och slutligen en ickelinjär transformering). Dessa metoder skiljer sig från varandra i hur de anpassar positioneringen av diffusionsbilden gentemot den högre upplösta bilden. I detta arbetet utvärderades även skillnaden i resultat när registreringarna använde diffusionsbilder med olika konstraster (olika b-värden) för att genomföra (driva) registreringarna. Kontrasterna som användes bestod av ett lågt b-värde (b0), vilket ger en bild med hög signal och därmed områden med hög intensitet. Ett högt b-värde (b2500) användes också, vilket ger en bild med lägre signal och mindre intensitetsskillnader i bilden. Utvärderingen av registreringsmetoderna har bestått av visuell evaluering samt evaluering av fördelningen av medeldiffusiviteten i hjärnbarken i form av PVE-index.

Resultatet av den visuella evalueringen visade att det fanns en skillnad mellan de olika undersökta metoderna för att utföra en registrering på. Resultaten av PVE-indexen visade på att combined presterade sämst utav metoderna som utvärderades. I fallen när b0 drev registreringarna så visade resultaten att rigid resulterade i det minsta PVE-indexet. När b2500 drev registreringarna så visade resultaten istället att registreringarna med transformerna rigid, affine och nonlinear presterade likvärdigt. Resultaten visade även på att det fanns en skillnad mellan när b0 och b2500 drev registreringarna. (Less)
Please use this url to cite or link to this publication:
author
Alm Ekwall, Julia
supervisor
organization
course
MSFT02 20241
year
type
H2 - Master's Degree (Two Years)
subject
language
English
id
9160334
date added to LUP
2024-06-20 08:34:56
date last changed
2024-06-20 19:22:48
@misc{9160334,
  abstract     = {{Introduction and aim
The changes in the brain caused by Alzheimer’s disease usually begin many years before the first symptoms arise. The effect of the neurodegenerative disease involves both morphological and microstructural changes in the brain, due to neurons dying. These changes in the microstructure can be measured with diffusion MRI. The resolution however limits the accuracy of the imaging, making structures more difficult to distinguish due to partial volume effects (PVE). PVE occurs when voxels contain information from multiple tissues, which takes place more frequently when the resolution is low. To reduce the PVE the diffusion-weighted image (DWI) can be registered to a T1-weighted image. The registration enables the use of a segmentation of the gray and white matter from the T1-weighted image. This segmentation can be used on the DWI to measure changes in the microstructure of the brain. There are however many ways to register images, therefore there is a need to identify which method of registration that produces the best alignment between images with the least influence of PVE. There are additional steps on how to reduce the PVE, such as using a Bowsher prior method where the registered diffusion images are reconstructed into a higher resolution.

Material and Methods
Registrations were performed in different ways by using the different transforms rigid, affine and nonlinear. A combined method with multiple input images was also used to perform a registration in this work. The registrations were alternately driven by a b0-image or a b2500-image. The b0 and b2500 are images with different contrasts, where the b0 has a low diffusion weighting and the b2500 has a high diffusion weighting. The registrations were evaluated by visual evaluations and by extracting mean diffusivity (MD) values from the cortex. The MD-values in the cortex range up to approximately 1.1 · 10−3 mm2/s for a cognitively normal individual. Values higher than this occur due to PVE with the cerebrospinal fluid (CSF) surrounding the cortex. The fraction of voxels in the cortex with an MD-value above 1.1 · 10−3 mm2/s was therefore used as a PVE-index. Furthermore, the distribution of PVE-indices was evaluated in 70 cortical regions which were extracted with the Desikan–Killiany atlas available in FreeSurfer. A reconstruction of the registered diffusion-weighted images was also performed with the Bowsher prior method.

Results
The registrations showed a notable difference in the visual evaluations, especially when the b0 was driving the registrations. From the evaluation of the PVE-index, the results indicated that it was superior to drive the registration with the b2500 instead of the b0. A paired t-test confirmed that there was a statistically significant difference between driving the registrations using either the b0 or the b2500. The PVE-indices varied between different registrations as well. The difference was more notable between registrations when the b0 was used in comparison to the b2500. A two-way ANOVA without reproducibility showed a statistically significant difference between registrations for both b0 and b2500 driving the registrations. The ANOVA was followed up with Tukey’s post hoc honestly significant difference (HSD) test. It showed a statistically significant difference between all registrations when the b0 was used. For when the b2500 was driving the registrations, all registrations showed a statistically significant difference except for between the rigid and affine and between the affine and nonlinear transforms. The reconstructed diffusion images with the Bowsher prior method resulted in a visually higher resolution. The algorithm however altered the values in the images which prevented further evaluation of this approach.

Conclusion
The results indicated that the rigid transform produced the highest registration accuracy when the b0 was used to drive the registrations. When the b2500 was used to drive the registrations, the accuracy of the registrations increased and the result suggested that the accuracy was similar for rigid, affine and nonlinear. The result also indicated that the analysis of registration accuracy is not straightforward.}},
  author       = {{Alm Ekwall, Julia}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Investigation of Image Co-Registration for Mapping Cortical Microstructures using dMRI}},
  year         = {{2024}},
}