Created at 6am, Mar 27
Ms-RAGScience
0
Predicting Parkinson’s disease trajectory using clinical and functional MRI features: a reproduction and replication study
Contract ID
TFwQuNHhEXMtkG1dE6C_RbIlBrIigtExMQh7kuijP5k
File Type
PDF
Entry Count
142
Embed. Model
jina_embeddings_v2_base_en
Index Type
hnsw

Elodie Germani1,, Nikhil Baghwat2, Mathieu Dugre3, Remi Gau2, Albert Montillo4, Kevin Nguyen4, Andrzej Soko lowski3, Madeleine Sharp2, Jean-Baptiste Poline2,+,Tristan Glatard3,+1- Univ Rennes, Inria, CNRS, Inserm, France2- Department of Neurology and Neurosurgery, McGill University, Montreal,Canada3- Department of Computer Science and Software Engineering, ConcordiaUniversity, Montreal, Canada4- Lyda Hill Department of Bioinformatics, University of Texas SouthwesternMedical Center, Dallas, USA+ Equal contributions elodie.germani@irisa.frAbstractParkinson’s disease (PD) is a common neurodegenerative disorder with a poorly understood physiopathology. In clinical practice, challenges are encountered in the diagnosis of early stages and in the prediction of the disease progression due to the absence of established biomarkers. Several biomarkers obtained using neuroimaging techniques such as functional Magnetic Resonance Imaging (fMRI) have been studied recently. However, the reliability and generalizability of neuroimaging-based measurements are susceptible to several different sources of variability, including those introduced by different analysis methods or population sampling. In this context, an evaluation of the robustness of such biomarkers is essential. This study is part of a larger project investigating the replicability of potential neuroimaging biomarkers of PD. Here, we attempt to reproduce (same data, same method) and replicate (different data or method) the models described in [1] to predict individual’s PD current state and progression using demographic, clinical and neuroimaging features (fALFF and ReHo extracted from resting-state fMRI). Weused the Parkinson’s Progression Markers Initiative dataset (PPMI, ppmi-info.org), as in [1] and tried to reproduce the original cohort, imaging features and machine learning models as closely as possible using the information available in the paper and the code. We also investigated methodological variations in cohort selection, feature extraction pipelines and sets of input features. Using the reproduction workflow, we managed to obtain better than chance performance for all our models (R2 > 0), but this performance remained very different from the ones reported in the original study. The challenges encountered while reproducing and replicating the original work are likely explained by the complexity of neuroimaging studies, in particular in clinical settings. We provide recommendations to facilitate the reproducibility of such studies in the future, for instance with the use of version control tools, standardization of pipelines and publication of analysis code and derived data.

Indeed, out of 102, 80 showed at least one time point with a frame-wise displacement superior to 0.5mm. However, since the authors in did not remove high-motion volumes within participants, and that completely removing participants with high-motion volumes would highly decrease our cohorts sample size, we chose to keep all participants and all volumes. Regarding segmentation masks, after visual inspection no significant artifact was found for any participants using AFNI segmentation in default workflow. For some participants, small distortions were found in particular close to brain extremities (interhemispheric area or close to the skull in occipital and parietal regions). Using FSL segmentation however, we found brain masking issues that had impacts on segmentation quality. We used BET using default parameters to skullstrip images before segmentation and since we chose to explore the impact of different default implementations of pipelines, we did not exclude the segmentations
id: 831217a9ef8aba16b81b19a33903107b - page: 16
Most participants of the study showed high movement parameters. With the fMRIprep pipeline, observations were similar regarding movement parameters and registration. There was no large artefact in the segmentation masks.
id: 523af237003f6324317d15bc0bdac1cf - page: 16
Performance of the default workflow The first objective of this study was to reproduce the models described in and to compare their performance with the one in the original study. In our default workflow, we implemented the default choices described in Figure 1: closest-to-original cohort, image pre-processing pipeline with AFNI segmentation, z-scoring of whole-brain fALFF and ReHo maps, use of all demographic, clinical and imaging features described in the original paper, and the model selection method derived from the authors code. We trained 12 models per time point (Baseline, Year 1, Year 2, Year 4) and imaging feature (fALFF or ReHo), corresponding to 4 machine learning models 3 brain parcellations. We reported for each imaging feature and time point the performance of the 12 models in Table 3.
id: 695b8bd0101993310a38747c00a41df4 - page: 16
Chance levels were computed using permutation tests as described in the Evaluation metrics section. We obtained R2 values that represented the chance prediction performance at different time point for fALFF and ReHo. These values are also presented in Table 3. Using the default workflow, we obtained different performance for all models parcellation combination. The best performance across these combination was different from those reported in at all time point. At Baseline, our best model performed better than chance but we did not manage to reach a R2 value close to the one reported in the original paper with any model or feature. Moreover, the best-performing models were 16/34 Figure 3. Distribution of MDS-UPDRS scores reported in the original papers cohort (top: Figure S1 extracted from ), the replication cohort (middle) and the closest-tooriginal cohort (bottom).
id: 6c48c17e036edaca9e72c2e2ea627cd6 - page: 16
How to Retrieve?
# Search

curl -X POST "https://search.dria.co/hnsw/search" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"rerank": true, "top_n": 10, "contract_id": "TFwQuNHhEXMtkG1dE6C_RbIlBrIigtExMQh7kuijP5k", "query": "What is alexanDRIA library?"}'
        
# Query

curl -X POST "https://search.dria.co/hnsw/query" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"vector": [0.123, 0.5236], "top_n": 10, "contract_id": "TFwQuNHhEXMtkG1dE6C_RbIlBrIigtExMQh7kuijP5k", "level": 2}'