TY - GEN
T1 - Diffusion MRI as a decision making tool for in-room MRI-guided radiotherapy
AU - Højmark Bisgaard, Anne Louise
PY - 2023/10/19
Y1 - 2023/10/19
N2 - Magnetic resonance imaging (MRI) plays an increasing role in radiotherapy (RT) as itprovides excellent soft tissue contrast. One of the most recent technical advances in RT isthe hybrid MRI linear accelerator (MRI-linac), an integrated MRI scanner and linearaccelerator, which enables adaptation of the treatment to match the patient’s anatomy-ofthe-day. In addition, functional MRI techniques such as diffusion-weighted MRI (DWI)provide biological information of body tissue, and hereby holds great potential forbiologically guided RT. DWI allows derivation of quantitative parameters such as the apparent diffusion coefficient (ADC), which may be used to identify radio-resistant regions within the tumour, and are potential biomarkers for response to RT. DWI might thus be used as a tool to guide treatment decisions in the clinic and hereby help personalizing the treatment to the individual patient.Before DWI can become a decision making tool in the clinic, a technical validation must be performed to ensure consistency of biomarker measurements in terms of inter- and intraobserver variation, repeatability and reproducibility. Further, a clinical validation is required to establish a relationship between biomarkers and clinical outcomes. With the overall goal of translating DWI into clinical use, this thesis addresses some of the challenges in relation to a technical and clinical validation of DWI. Study 1: Delineation of regions of interest for ADC measurementsADC measurements require delineation of a region of interest (ROI), which is observer
dependent and time consuming if performed manually. Study 1 presented a simple,
threshold-based semi-automatic delineation tool for ADC measurements, and evaluated its
performance in patients with rectal cancer. Compared with manual delineation, the tool
showed a slightly smaller intra-observer ADC variation. Moreover, it was capable of
detecting temporal changes in ADC larger than the repeatability, indicating that the
observed ADC changes are true changes and not only a consequence of measurement
uncertainty. Thus, the tool may become useful for a technical validation of ADC. Study 2: Multi-centre ADC reproducibility Varying methods for ADC derivation between centres (i.e. hospitals or research
institutions) lead to a poor multi-centre ADC reproducibility and hamper the technical
validation of ADC. Study 2 evaluated the ADC variation between nine MRI-linac centres
with respect to two categories: delineation and calculation. MRI scans were shared
between the centres, and each centre performed 1) delineation of ROIs and 2) ADC
calculation within the delineated ROIs from all centres using the centre’s local calculation
method. A correlation between delineation variation and ADC variation was observed,
underlining the need for and importance of a reduction of the delineation uncertainty.
Interestingly, the calculation-related ADC variation was even larger than delineationrelated ADC variation, and was mainly driven by different choices of b-values. Other
important factors were the software implementation and whether the calculation was
performed voxel-wise or based on the mean DWI signal within the ROI. It is the hope that
these findings will be useful when designing future studies of ADC as a biomarker,
especially in a multi-centre setting. Study 3: Prediction of response using longitudinal DWI The MRI-linac has made daily acquisition of DWI feasible, however, the literature
investigating the relationship between longitudinal DWI acquired during the RT course
and clinical outcome is sparse. Study 3 investigated the value of longitudinal DWI for
prediction of overall survival in patients with pancreatic cancer using both a standard,
model-based method (ADC) and a model-free decomposition method for parameter
derivation. The best prediction model for overall survival based on cross-validation
included two DWI parameters, both derived using decomposition of the DWI signal.
Moreover, the best model included both baseline information and DWI changes during the
RT course, and hence demonstrated potential value of longitudinal DWI for response
prediction.ConclusionsThis thesis has addressed important steps in both technical and clinical validation of DWI.
It was shown that semi-automatic delineation using a could be a way to improve
consistency of ADC measurements and save time compared to manual delineation.
Moreover, recommendations for improved multi-centre reproducibility of ADC were
provided. These findings can be seen as initiatives to improve the repeatability and
reproducibility of ADC measurements, and might be useful in future validation studies of ADC as a response biomarker. Furthermore, the prognostic value of longitudinal DWI for
prediction of overall survival in patients with pancreatic cancer was demonstrated,
indicating that longitudinal DWI could potentially be useful for prediction of response. Perspectives
AB - Magnetic resonance imaging (MRI) plays an increasing role in radiotherapy (RT) as itprovides excellent soft tissue contrast. One of the most recent technical advances in RT isthe hybrid MRI linear accelerator (MRI-linac), an integrated MRI scanner and linearaccelerator, which enables adaptation of the treatment to match the patient’s anatomy-ofthe-day. In addition, functional MRI techniques such as diffusion-weighted MRI (DWI)provide biological information of body tissue, and hereby holds great potential forbiologically guided RT. DWI allows derivation of quantitative parameters such as the apparent diffusion coefficient (ADC), which may be used to identify radio-resistant regions within the tumour, and are potential biomarkers for response to RT. DWI might thus be used as a tool to guide treatment decisions in the clinic and hereby help personalizing the treatment to the individual patient.Before DWI can become a decision making tool in the clinic, a technical validation must be performed to ensure consistency of biomarker measurements in terms of inter- and intraobserver variation, repeatability and reproducibility. Further, a clinical validation is required to establish a relationship between biomarkers and clinical outcomes. With the overall goal of translating DWI into clinical use, this thesis addresses some of the challenges in relation to a technical and clinical validation of DWI. Study 1: Delineation of regions of interest for ADC measurementsADC measurements require delineation of a region of interest (ROI), which is observer
dependent and time consuming if performed manually. Study 1 presented a simple,
threshold-based semi-automatic delineation tool for ADC measurements, and evaluated its
performance in patients with rectal cancer. Compared with manual delineation, the tool
showed a slightly smaller intra-observer ADC variation. Moreover, it was capable of
detecting temporal changes in ADC larger than the repeatability, indicating that the
observed ADC changes are true changes and not only a consequence of measurement
uncertainty. Thus, the tool may become useful for a technical validation of ADC. Study 2: Multi-centre ADC reproducibility Varying methods for ADC derivation between centres (i.e. hospitals or research
institutions) lead to a poor multi-centre ADC reproducibility and hamper the technical
validation of ADC. Study 2 evaluated the ADC variation between nine MRI-linac centres
with respect to two categories: delineation and calculation. MRI scans were shared
between the centres, and each centre performed 1) delineation of ROIs and 2) ADC
calculation within the delineated ROIs from all centres using the centre’s local calculation
method. A correlation between delineation variation and ADC variation was observed,
underlining the need for and importance of a reduction of the delineation uncertainty.
Interestingly, the calculation-related ADC variation was even larger than delineationrelated ADC variation, and was mainly driven by different choices of b-values. Other
important factors were the software implementation and whether the calculation was
performed voxel-wise or based on the mean DWI signal within the ROI. It is the hope that
these findings will be useful when designing future studies of ADC as a biomarker,
especially in a multi-centre setting. Study 3: Prediction of response using longitudinal DWI The MRI-linac has made daily acquisition of DWI feasible, however, the literature
investigating the relationship between longitudinal DWI acquired during the RT course
and clinical outcome is sparse. Study 3 investigated the value of longitudinal DWI for
prediction of overall survival in patients with pancreatic cancer using both a standard,
model-based method (ADC) and a model-free decomposition method for parameter
derivation. The best prediction model for overall survival based on cross-validation
included two DWI parameters, both derived using decomposition of the DWI signal.
Moreover, the best model included both baseline information and DWI changes during the
RT course, and hence demonstrated potential value of longitudinal DWI for response
prediction.ConclusionsThis thesis has addressed important steps in both technical and clinical validation of DWI.
It was shown that semi-automatic delineation using a could be a way to improve
consistency of ADC measurements and save time compared to manual delineation.
Moreover, recommendations for improved multi-centre reproducibility of ADC were
provided. These findings can be seen as initiatives to improve the repeatability and
reproducibility of ADC measurements, and might be useful in future validation studies of ADC as a response biomarker. Furthermore, the prognostic value of longitudinal DWI for
prediction of overall survival in patients with pancreatic cancer was demonstrated,
indicating that longitudinal DWI could potentially be useful for prediction of response. Perspectives
KW - Magnetisk resonans billeddannelse
KW - Diffusions-vægtet MR
KW - Apparent diffusion coefficient
KW - biomarkører
KW - MRI-linac
KW - MR-vejledt stråleterapi
KW - Magnetic resonance imaging
KW - Diffusion-weighted MRI
KW - Apparent diffusion coefficient
KW - Quantitative imaging biomarkers
KW - MRI-linac
KW - MRI-guided radiotherapy
U2 - 10.21996/228s-xa44
DO - 10.21996/228s-xa44
M3 - Ph.D. thesis
PB - Syddansk Universitet. Det Sundhedsvidenskabelige Fakultet
ER -