{"id":{"repo_id":"calgary","oai_identifier":"oai:ucalgary.scholaris.ca:1880/122394"},"canonical_url":"https://search.dev.ndltd.org/etd/calgary/oai:ucalgary.scholaris.ca:1880/122394","repository":{"repo_id":"calgary","name":"University of Calgary","base_url":"https://ucalgary.scholaris.ca/server/oai/request"},"display":{"title":"Machine Learning Methods for Personalized Treatment Response Characterization Using Clinical Care Brain MRI and Non-Imaging Data in Multiple Sclerosis","abstract":"Multiple sclerosis (MS) is a common and disabling neurological disease with increasing cost. Most people begin MS with a relapsing-remitting form (RRMS) but disease trajectory differs by person. There are over 20 low- and high-efficacy disease modifying therapies (DMTs) for people with RRMS; however, many of them still experience worsening disease. The inability to optimally personalize treatment is core. Deep learning methods serve as ideal candidates to handle person-specific disease outcomes. This thesis aimed to investigate deep learning models for predicting treatment response in RRMS individuals 2 years after starting or switching to a new DMT using baseline clinical data, especially brain magnetic resonance imaging (MRI). Two RRMS cohorts (2/3 women) at 169 and 75 participants were examined, used for training and semi-external testing, respectively. Six key clinical variables and common brain MRI sequences: T1-weighted, T2-weighted, and FLAIR were assessed. The first Aim was dealing with ‘missing’ MRI sequences as commonly seen in clinical care data using a leading deep learning technique called CycleGAN. The focus was on T1-weighted MRI that was unavailable the most. The second Aim was to investigate models based on a top-ranking convolutional neural network, ResNet50, for predicting 2-year treatment response using Clinical Only, MRI Only, or Combined input. To handle class imbalance, this study applied a unique dural-labeling system between training and testing. Overall results showed the robustness of CycleGAN for image synthesis, thebest with synthesizing T1-weighted MRI from T2-weighted images. Using ResNet50, the MRI Only model performed the best with a high average AUC of 0.88 and 0.74 over internal and semi-external testing. The Combined model performed similarly. Further exploration of the best MRI Only model suggested that the prediction was better in predicting persons starting than switching to a new DMT, and prediction using CycleGAN-synthesized images performed similarly to using source images. Overall findings indicate the feasibility of using routine clinical brain MRI data along with deep learning for estimating 2-year treatment response in RRMS.","abstract_html":"Multiple sclerosis (MS) is a common and disabling neurological disease with increasing cost. Most people begin MS with a relapsing-remitting form (RRMS) but disease trajectory differs by person. There are over 20 low- and high-efficacy disease modifying therapies (DMTs) for people with RRMS; however, many of them still experience worsening disease. The inability to optimally personalize treatment is core. Deep learning methods serve as ideal candidates to handle person-specific disease outcomes. This thesis aimed to investigate deep learning models for predicting treatment response in RRMS individuals 2 years after starting or switching to a new DMT using baseline clinical data, especially brain magnetic resonance imaging (MRI). Two RRMS cohorts (2/3 women) at 169 and 75 participants were examined, used for training and semi-external testing, respectively. Six key clinical variables and common brain MRI sequences: T1-weighted, T2-weighted, and FLAIR were assessed. The first Aim was dealing with ‘missing’ MRI sequences as commonly seen in clinical care data using a leading deep learning technique called CycleGAN. The focus was on T1-weighted MRI that was unavailable the most. The second Aim was to investigate models based on a top-ranking convolutional neural network, ResNet50, for predicting 2-year treatment response using Clinical Only, MRI Only, or Combined input. To handle class imbalance, this study applied a unique dural-labeling system between training and testing. Overall results showed the robustness of CycleGAN for image synthesis, thebest with synthesizing T1-weighted MRI from T2-weighted images. Using ResNet50, the MRI Only model performed the best with a high average AUC of 0.88 and 0.74 over internal and semi-external testing. The Combined model performed similarly. Further exploration of the best MRI Only model suggested that the prediction was better in predicting persons starting than switching to a new DMT, and prediction using CycleGAN-synthesized images performed similarly to using source images. Overall findings indicate the feasibility of using routine clinical brain MRI data along with deep learning for estimating 2-year treatment response in RRMS.","abstract_has_math":false,"creators":["Tariq, Rehman"],"institution":"Schulich School of Engineering","degree_name":"Master of Science (MSc)","degree_level":null,"degree_discipline":"Engineering – Biomedical","degree_department":null,"school":null,"contributors":[],"advisors":["Zhang, Yunyan"],"committee_chairs":[],"committee_members":["Bento, Mariana","Cámara-Lemarroy, Carlos"],"year":2025,"date_issued":"2025-07-14","date_published":"2025-07-14","updated_at":"2026-07-24T01:30:38Z","subjects":["Brain MRI","Transfer Learning","Multiple Sclerosis"],"languages":["en"],"rights":["Unless otherwise indicated, this material is protected by copyright and has been made available with authorization from the copyright owner. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission."],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://dx.doi.org/10.11575/PRISM/49987"],"render_values":[{"text":"https://dx.doi.org/10.11575/PRISM/49987","href":"https://dx.doi.org/10.11575/PRISM/49987","code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/1880/122394","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Zhang, Yunyan"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Bento, Mariana","Cámara-Lemarroy, Carlos"]},{"key":"dc:creator","label":"Author","values":["Tariq, Rehman"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025-11"]},{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-09-16T17:01:47Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-09-16T17:01:47Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-07-14"]},{"key":"dc:type","label":"Dc Type","values":["master thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Engineering – Biomedical"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science (MSc)"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Calgary"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Brain MRI","Transfer Learning","Multiple Sclerosis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Unless otherwise indicated, this material is protected by copyright and has been made available with authorization from the copyright owner. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission."]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://dx.doi.org/10.11575/PRISM/49987"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/1880/122394"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Multiple sclerosis (MS) is a common and disabling neurological disease with increasing cost. Most people begin MS with a relapsing-remitting form (RRMS) but disease trajectory differs by person. There are over 20 low- and high-efficacy disease modifying therapies (DMTs) for people with RRMS; however, many of them still experience worsening disease. The inability to optimally personalize treatment is core. Deep learning methods serve as ideal candidates to handle person-specific disease outcomes. This thesis aimed to investigate deep learning models for predicting treatment response in RRMS individuals 2 years after starting or switching to a new DMT using baseline clinical data, especially brain magnetic resonance imaging (MRI). Two RRMS cohorts (2/3 women) at 169 and 75 participants were examined, used for training and semi-external testing, respectively. Six key clinical variables and common brain MRI sequences: T1-weighted, T2-weighted, and FLAIR were assessed. The first Aim was dealing with ‘missing’ MRI sequences as commonly seen in clinical care data using a leading deep learning technique called CycleGAN. The focus was on T1-weighted MRI that was unavailable the most. The second Aim was to investigate models based on a top-ranking convolutional neural network, ResNet50, for predicting 2-year treatment response using Clinical Only, MRI Only, or Combined input. To handle class imbalance, this study applied a unique dural-labeling system between training and testing. Overall results showed the robustness of CycleGAN for image synthesis, thebest with synthesizing T1-weighted MRI from T2-weighted images. Using ResNet50, the MRI Only model performed the best with a high average AUC of 0.88 and 0.74 over internal and semi-external testing. The Combined model performed similarly. Further exploration of the best MRI Only model suggested that the prediction was better in predicting persons starting than switching to a new DMT, and prediction using CycleGAN-synthesized images performed similarly to using source images. Overall findings indicate the feasibility of using routine clinical brain MRI data along with deep learning for estimating 2-year treatment response in RRMS."]},{"key":"dc:title","label":"Title","values":["Machine Learning Methods for Personalized Treatment Response Characterization Using Clinical Care Brain MRI and Non-Imaging Data in Multiple Sclerosis"]}]}],"canonical_facts":{"dc:contributor.advisor":["Zhang, Yunyan"],"dc:contributor.committeemember":["Bento, Mariana","Cámara-Lemarroy, Carlos"],"dc:creator":["Tariq, Rehman"],"dc:date":["2025-11"],"dc:date.accessioned":["2025-09-16T17:01:47Z"],"dc:date.available":["2025-09-16T17:01:47Z"],"dc:date.issued":["2025-07-14"],"dc:description.abstract":["Multiple sclerosis (MS) is a common and disabling neurological disease with increasing cost. Most people begin MS with a relapsing-remitting form (RRMS) but disease trajectory differs by person. There are over 20 low- and high-efficacy disease modifying therapies (DMTs) for people with RRMS; however, many of them still experience worsening disease. The inability to optimally personalize treatment is core. Deep learning methods serve as ideal candidates to handle person-specific disease outcomes. This thesis aimed to investigate deep learning models for predicting treatment response in RRMS individuals 2 years after starting or switching to a new DMT using baseline clinical data, especially brain magnetic resonance imaging (MRI). Two RRMS cohorts (2/3 women) at 169 and 75 participants were examined, used for training and semi-external testing, respectively. Six key clinical variables and common brain MRI sequences: T1-weighted, T2-weighted, and FLAIR were assessed. The first Aim was dealing with ‘missing’ MRI sequences as commonly seen in clinical care data using a leading deep learning technique called CycleGAN. The focus was on T1-weighted MRI that was unavailable the most. The second Aim was to investigate models based on a top-ranking convolutional neural network, ResNet50, for predicting 2-year treatment response using Clinical Only, MRI Only, or Combined input. To handle class imbalance, this study applied a unique dural-labeling system between training and testing. Overall results showed the robustness of CycleGAN for image synthesis, thebest with synthesizing T1-weighted MRI from T2-weighted images. Using ResNet50, the MRI Only model performed the best with a high average AUC of 0.88 and 0.74 over internal and semi-external testing. The Combined model performed similarly. Further exploration of the best MRI Only model suggested that the prediction was better in predicting persons starting than switching to a new DMT, and prediction using CycleGAN-synthesized images performed similarly to using source images. Overall findings indicate the feasibility of using routine clinical brain MRI data along with deep learning for estimating 2-year treatment response in RRMS."],"dc:identifier.doi":["https://dx.doi.org/10.11575/PRISM/49987"],"dc:identifier.uri":["https://hdl.handle.net/1880/122394"],"dc:language.iso":["en"],"dc:rights":["Unless otherwise indicated, this material is protected by copyright and has been made available with authorization from the copyright owner. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission."],"dc:subject":["Brain MRI","Transfer Learning","Multiple Sclerosis"],"dc:title":["Machine Learning Methods for Personalized Treatment Response Characterization Using Clinical Care Brain MRI and Non-Imaging Data in Multiple Sclerosis"],"dc:type":["master thesis"],"thesis:degree_discipline":["Engineering – Biomedical"],"thesis:degree_name":["Master of Science (MSc)"],"thesis:institution_name":["University of Calgary"]},"updated_at":"2026-07-24T01:30:38Z"}