EEG-GCNN: Dataset, Task, Models, and Neurological Disease Detection Pipeline

.gitignore

@@ -76,6 +76,7 @@ instance/
 
 # Sphinx documentation
 docs/_build/
+docs/**/_autosummary/
 
 # PyBuilder
 target/

docs/api/datasets.rst

@@ -230,6 +230,7 @@ Available Datasets
     datasets/pyhealth.datasets.ISRUCDataset
     datasets/pyhealth.datasets.MIMICExtractDataset
     datasets/pyhealth.datasets.OMOPDataset
+    datasets/pyhealth.datasets.EEGGCNNRawDataset
     datasets/pyhealth.datasets.DREAMTDataset
     datasets/pyhealth.datasets.SHHSDataset
     datasets/pyhealth.datasets.SleepEDFDataset
@@ -241,6 +242,7 @@ Available Datasets
     datasets/pyhealth.datasets.PhysioNetDeIDDataset
     datasets/pyhealth.datasets.TUABDataset
     datasets/pyhealth.datasets.TUEVDataset
+    datasets/pyhealth.datasets.EEGGCNNDataset
     datasets/pyhealth.datasets.ClinVarDataset
     datasets/pyhealth.datasets.COSMICDataset
     datasets/pyhealth.datasets.TCGAPRADDataset

docs/api/datasets/pyhealth.datasets.EEGGCNNDataset.rst

@@ -0,0 +1,30 @@
+pyhealth.datasets.EEGGCNNDataset
+=================================
+
+Dataset is available at https://figshare.com/articles/dataset/EEG-GCNN_Augmenting_Electroencephalogram-based_Neurological_Disease_Diagnosis_using_a_Domain-guided_Graph_Convolutional_Neural_Network/13251509
+
+The EEG-GCNN dataset is derived from the pre-computed features introduced in:
+
+    Wagh, N., & Varatharajah, Y. (2020). EEG-GCNN: Augmenting
+    Electroencephalogram-based Neurological Disease Diagnosis using a
+    Domain-guided Graph Convolutional Neural Network.
+    *Machine Learning for Health (ML4H) workshop, NeurIPS 2020*.
+    https://arxiv.org/abs/2011.10432
+
+Each EEG recording is segmented into fixed-length windows. Every window is
+represented as a fully-connected graph over 8 bipolar EEG electrode pairs.
+Node features are six PSD band-power values (delta, theta, alpha, beta,
+low-gamma, high-gamma) per electrode. Edge weights encode a blended
+connectivity measure combining geodesic distance between electrode positions
+and spectral coherence between electrode signals, controlled by a mixing
+parameter ``alpha``.
+
+The dataset supports neurological disease classification (diseased vs.
+healthy) and is designed for use with graph neural network models such as
+:class:`~pyhealth.models.EEGGraphConvNet` and
+:class:`~pyhealth.models.EEGGATConvNet`.
+
+.. autoclass:: pyhealth.datasets.EEGGCNNDataset
+    :members:
+    :undoc-members:
+    :show-inheritance:

docs/api/datasets/pyhealth.datasets.EEGGCNNRawDataset.rst

@@ -0,0 +1,16 @@
+pyhealth.datasets.EEGGCNNRawDataset
+======================================
+
+Raw EEG dataset for the EEG-GCNN paper (Wagh & Varatharajah, ML4H @ NeurIPS 2020).
+Processes raw EDF/BrainVision files from TUAB normal-subset (patient EEGs) and
+MPI LEMON (healthy controls).
+
+See also :class:`~pyhealth.datasets.EEGGCNNDataset` for the pre-computed
+FigShare feature variant.
+
+Paper: https://proceedings.mlr.press/v136/wagh20a.html
+
+.. autoclass:: pyhealth.datasets.EEGGCNNRawDataset
+    :members:
+    :undoc-members:
+    :show-inheritance: