diff --git a/_projects/federated-learning.md b/_projects/federated-learning.md new file mode 100644 index 00000000000..945357f8f1a --- /dev/null +++ b/_projects/federated-learning.md @@ -0,0 +1,22 @@ +--- +title: Benchmarking federated learning approaches against siloed and mega-analysis regimes +# subtitle: +status: active +image: img/projects/fl_overview.png +layout: project +people: + - Michelle + - Nikhil +--- + +Although neuroimaging is seeing a growing number of datasets, the international adoption of strong data privacy frameworks ([Marelli & Testa, 2018](https://doi.org/10.1126/science.aar5419)) has led to many of these datasets remaining in so-called “silos”. When data cannot readily be shared, it becomes imperative to develop distributed data processing tools and federated analysis methods to enable large-scale multi-site studies. + +In this project, we compare a simple federated analysis setup (i.e. sharing only fitted model parameters) with two traditional experimental setups: +- Siloed analysis (no sharing of data or model parameters) +- Mega-analysis (sharing data) + +We evaluate the performance of machine learning (ML) models on several neuroimaging datasets of Parkinson’s (PD) and Alzheimer’s disease (AD) on two common prediction tasks in neurodegenerative diseases: 1) brain age and 2) cognitive decline. We hypothesize that model performance improves as we go from siloed to federated to mega-analysis setups. + +

+ Machine learning setups +

diff --git a/_projects/nipoppy.md b/_projects/nipoppy.md index 1093de7c4ad..987e1f67752 100644 --- a/_projects/nipoppy.md +++ b/_projects/nipoppy.md @@ -1,15 +1,19 @@ --- title: "Nipoppy: A framework for the reproducible organization and processing neuroimaging-clinical datasets" subtitle: A Neurobagel complement project -image: img/other/nipoppy_overview.png +image: img/projects/nipoppy_protocol.png status: active layout: project -repository: https://github.com/neurodatascience/nipoppy/tree/main -documentation: https://www.neurobagel.org/documentation/Nipoppy/overview/ +repository: https://github.com/nipoppy/nipoppy/tree/main +documentation: https://nipoppy.readthedocs.io/en/latest people: - Michelle - Nikhil + - Mathieu - Brent + - Mohammad + - Jacob + - Julia - Remi - Vincent - Ines @@ -21,24 +25,25 @@ Nipoppy is a lightweight framework for standardized organization and processing The framework consists of the following: -1. A **data organization specification** for both imaging and non-imaging data - - - Whenever possible, we follow the [Brain Imaging Data Structure (BIDS) standard](https://bids.neuroimaging.io/) - - We provide additional specifications for cases not (yet) covered by BIDS, such as phenotypic data and imaging derivatives - -

- Nipoppy dataset layout -

- -2. A **standardized workflow process (i.e., protocol)** covering the following: +1. A **standardized workflow process (i.e., protocol)** covering the following: - Curation and organization of MRI and tabular data - E.g., conversion of DICOM data to BIDS - Standardized processing of imaging data using existing or custom pipelines - - E.g., fMRIPrep, MRIQC + - E.g., [fMRIPrep](https://fmriprep.org/en/stable/), [MRIQC](https://mriqc.readthedocs.io/en/stable/) - The [Boutiques framework](https://boutiques.github.io/) is used to flexibly execute and add new pipelines - Tracking of availability status for raw and processed data - We also develop an interactive [dashboard](https://digest.neurobagel.org/) for easy visualization of tracker results - Extraction of imaging-derived phenotypes (IDPs) into files ready for downstream analysis -3. A [software package](https://github.com/neurodatascience/nipoppy/tree/main) with tools to help work within the framework + +2. A **data organization specification** for both imaging and non-imaging data + + - Whenever possible, we follow the [Brain Imaging Data Structure (BIDS) standard](https://bids.neuroimaging.io/) + - We provide additional specifications for cases not (yet) covered by BIDS, such as phenotypic data and imaging derivatives + +

+ Nipoppy dataset layout +

+ +3. A [software package](https://github.com/nipoppy/nipoppy/tree/main) with tools to help work within the framework Nipoppy can also provide metadata to [Neurobagel](https://www.neurobagel.org/documentation/) tools to allow for data harmonization and federated search of participants across multiple studies. diff --git a/img/other/nipoppy_layout.jpg b/img/other/nipoppy_layout.jpg deleted file mode 100644 index f60160b772d..00000000000 Binary files a/img/other/nipoppy_layout.jpg and /dev/null differ diff --git a/img/other/nipoppy_overview.png b/img/other/nipoppy_overview.png deleted file mode 100644 index e58adc370ee..00000000000 Binary files a/img/other/nipoppy_overview.png and /dev/null differ diff --git a/img/projects/fl_overview.png b/img/projects/fl_overview.png new file mode 100644 index 00000000000..0e21e041932 Binary files /dev/null and b/img/projects/fl_overview.png differ diff --git a/img/projects/fl_setups.jpg b/img/projects/fl_setups.jpg new file mode 100644 index 00000000000..10bef38d273 Binary files /dev/null and b/img/projects/fl_setups.jpg differ diff --git a/img/projects/nipoppy_protocol.jpg b/img/projects/nipoppy_protocol.jpg new file mode 100644 index 00000000000..0a6f5637ddb Binary files /dev/null and b/img/projects/nipoppy_protocol.jpg differ diff --git a/img/projects/nipoppy_specification.jpg b/img/projects/nipoppy_specification.jpg new file mode 100644 index 00000000000..7d0f8ebd9d2 Binary files /dev/null and b/img/projects/nipoppy_specification.jpg differ