Amelie Metz
@ameliemetz.bsky.social
🧠 PhD Candidate at McGill University and Douglas Research Center | frontotemporal dementia and neuroimaging | 🥨@🇨🇦
Reposted by Amelie Metz
🧾 Poster #487
Data-Driven Characterization of Heterogeneous Brain Atrophy and White Matter Hyperintensity Progression in FTD (Amelie Metz @ameliemetz.bsky.social )
Tuesday, July 29
Data-Driven Characterization of Heterogeneous Brain Atrophy and White Matter Hyperintensity Progression in FTD (Amelie Metz @ameliemetz.bsky.social )
Tuesday, July 29
July 22, 2025 at 10:35 PM
🧾 Poster #487
Data-Driven Characterization of Heterogeneous Brain Atrophy and White Matter Hyperintensity Progression in FTD (Amelie Metz @ameliemetz.bsky.social )
Tuesday, July 29
Data-Driven Characterization of Heterogeneous Brain Atrophy and White Matter Hyperintensity Progression in FTD (Amelie Metz @ameliemetz.bsky.social )
Tuesday, July 29
10/ Big THANK YOU to my advisors @dadarmahsa.bsky.social and @sylviavilleneuve.bsky.social and everyone who contributed to this work (aka @yasharneuro.bsky.social
and Simon Ducharme) for the ideas, feedback, and encouragement that made this possible!
Make sure to check out the full paper 😄
and Simon Ducharme) for the ideas, feedback, and encouragement that made this possible!
Make sure to check out the full paper 😄
February 24, 2025 at 9:14 PM
10/ Big THANK YOU to my advisors @dadarmahsa.bsky.social and @sylviavilleneuve.bsky.social and everyone who contributed to this work (aka @yasharneuro.bsky.social
and Simon Ducharme) for the ideas, feedback, and encouragement that made this possible!
Make sure to check out the full paper 😄
and Simon Ducharme) for the ideas, feedback, and encouragement that made this possible!
Make sure to check out the full paper 😄
9/ TAKEAWAY
🧠We can use the mapping between neurodegeneration and cognitive manifestations of the core FTD subtypes for phenotyping patients according to clinical variants.
🧠The inclusion of DBM measures adds to the classification precision in the absence of extensive clinical testing.
🧠We can use the mapping between neurodegeneration and cognitive manifestations of the core FTD subtypes for phenotyping patients according to clinical variants.
🧠The inclusion of DBM measures adds to the classification precision in the absence of extensive clinical testing.
February 24, 2025 at 9:14 PM
9/ TAKEAWAY
🧠We can use the mapping between neurodegeneration and cognitive manifestations of the core FTD subtypes for phenotyping patients according to clinical variants.
🧠The inclusion of DBM measures adds to the classification precision in the absence of extensive clinical testing.
🧠We can use the mapping between neurodegeneration and cognitive manifestations of the core FTD subtypes for phenotyping patients according to clinical variants.
🧠The inclusion of DBM measures adds to the classification precision in the absence of extensive clinical testing.
8/ To test the reliability of our approach, we validated our model in subsets of the data that were matched for disease severity, age, and scanner type, as well as in the longitudinal data of the same cohort, with similar results.
February 24, 2025 at 9:14 PM
8/ To test the reliability of our approach, we validated our model in subsets of the data that were matched for disease severity, age, and scanner type, as well as in the longitudinal data of the same cohort, with similar results.
7/ With minimal clinical input, the combination of atrophy and clinical measures was crucial. Using only the two cognitive scores in the classification, accuracy was reduced to 76.38% but notably, sensitivity and balanced accuracies highly increased when MRI values were added to the model.
February 24, 2025 at 9:14 PM
7/ With minimal clinical input, the combination of atrophy and clinical measures was crucial. Using only the two cognitive scores in the classification, accuracy was reduced to 76.38% but notably, sensitivity and balanced accuracies highly increased when MRI values were added to the model.
6/ To assess whether our model was adaptable for a clinical setting, we repeated our analyses while only including the CDR scales and one language test in the PLS and our classification model. Using these minimal variables, our model still achieved an accurate FTD subtype classification of 83.62!
February 24, 2025 at 9:14 PM
6/ To assess whether our model was adaptable for a clinical setting, we repeated our analyses while only including the CDR scales and one language test in the PLS and our classification model. Using these minimal variables, our model still achieved an accurate FTD subtype classification of 83.62!
5/ We ran our machine learning classifier with a 10-fold cross-validation loop on 100 randomized train and test splits, with PLS-based brain and cognition scores (including 16 cognitive scores) as inputs. The resulting three-class mean prediction accuracy over all repetitions was ‼️89.12%‼️
February 24, 2025 at 9:14 PM
5/ We ran our machine learning classifier with a 10-fold cross-validation loop on 100 randomized train and test splits, with PLS-based brain and cognition scores (including 16 cognitive scores) as inputs. The resulting three-class mean prediction accuracy over all repetitions was ‼️89.12%‼️
4/ Linear regression models assessing the relationship between atrophy and cognition patterns in each participant confirmed that the association between the two differs between FTD subtypes (bvFTD, svPPA, and nfvPPA), allowing us to use these profiles in a machine learning classification algorithm 👍
February 24, 2025 at 9:14 PM
4/ Linear regression models assessing the relationship between atrophy and cognition patterns in each participant confirmed that the association between the two differs between FTD subtypes (bvFTD, svPPA, and nfvPPA), allowing us to use these profiles in a machine learning classification algorithm 👍
3/ PLS results highlighted the involvement of frontal and temporal lobes and subcortical structures and linked atrophy in these regions to a decline in global cognitive function, language, and executive function.
February 24, 2025 at 9:14 PM
3/ PLS results highlighted the involvement of frontal and temporal lobes and subcortical structures and linked atrophy in these regions to a decline in global cognitive function, language, and executive function.
2/ We analyzed MRI and cognitive data from 136 participants in the NIFD cohort. (Sub)cortical atrophy 🧠was assessed using deformation-based morphometry. We then applied partial least squares, using resulting atrophy/cognition patterns to determine the most likely FTD subtype of each participant.
February 24, 2025 at 9:14 PM
2/ We analyzed MRI and cognitive data from 136 participants in the NIFD cohort. (Sub)cortical atrophy 🧠was assessed using deformation-based morphometry. We then applied partial least squares, using resulting atrophy/cognition patterns to determine the most likely FTD subtype of each participant.