| """EEG Annotation Tool adapter for SenuaLab EEG-DINO Medium.""" |
|
|
| from __future__ import annotations |
|
|
| try: |
| from braindecode.models import EEGDINO |
| except ImportError as exc: |
| raise ImportError("This model requires braindecode[hub]==1.6.1") from exc |
|
|
| from .preprocessing import INTERNATIONAL_10_20_CHANNELS, preprocess_batch as _preprocess |
|
|
|
|
| MODEL_CHANNELS = INTERNATIONAL_10_20_CHANNELS |
| MODEL_INPUT_SAMPLES = 800 |
| MODEL_SAMPLING_RATE_HZ = 200.0 |
| MODEL_WINDOW_SECONDS = 4.0 |
| MODEL_ENTRY_CLASS = "SenuaEEGDINOMedium" |
| MODEL_NUM_CLASSES = 2 |
| MODEL_CLASS_LABELS = ["Non-IED", "IED"] |
| MODEL_NON_IED_CLASS_INDEX = 0 |
| MODEL_IED_CLASS_INDICES = [1] |
| MODEL_DESCRIPTION = "SenuaLab partially fine-tuned EEG-DINO Medium binary IED detector" |
| MODEL_BATCH_PREPROCESSOR = "preprocess_batch" |
| MODEL_REQUIRED_REFERENCE = "common average (applied by model adapter)" |
| MODEL_REQUIRED_FILTERS = ["1-45 Hz zero-phase Butterworth (applied by model adapter)"] |
| MODEL_REQUIRED_NORMALIZATION = "global four-second window z-score, clipped to [-8,8]" |
| MODEL_INPUT_UNIT = "scale-invariant after window z-score" |
| MODEL_SOURCE_SIGNAL_POLICY = "raw" |
| MODEL_REQUIRES_FULL_WINDOW = True |
| MODEL_REQUIRES_ALL_CHANNELS = True |
| MODEL_DEFAULT_THRESHOLD = 0.619140625 |
| MODEL_DEFAULT_STEP_MS = 500.0 |
| MODEL_DEFAULT_PAD_POLICY = "skip" |
| MODEL_DEFAULT_BATCH_SIZE = 8 |
| MODEL_DEFAULT_BATCH_MEMORY_MB = 384.0 |
| MODEL_VALIDATION_NOTE = "Threshold selected on vEpiSet validation subjects; requires Braindecode 1.6.1 and is not externally validated." |
| MODEL_PREPROCESSING_NOTE = "The adapter applies the released 1-45 Hz, common-average, 200 Hz resampling, and global-window z-score pipeline." |
|
|
|
|
| def preprocess_batch(batch, source_sfreq=None, channel_names=None): |
| return _preprocess( |
| batch, |
| source_sfreq=source_sfreq, |
| target_sfreq=200, |
| target_samples=800, |
| channel_names=channel_names, |
| ) |
|
|
|
|
| class SenuaEEGDINOMedium(EEGDINO): |
| def __init__(self, num_channels: int = 19, num_classes: int = 2, input_length: int = 800): |
| if num_channels != 19 or input_length != 800: |
| raise ValueError("SenuaEEGDINOMedium requires 19 channels and 800 samples") |
| super().__init__( |
| n_outputs=num_classes, |
| n_chans=19, |
| n_times=800, |
| sfreq=200, |
| patch_size=200, |
| n_layer=16, |
| nhead=8, |
| dim_feedforward=1024, |
| channels_kernel_stride_padding_norm=( |
| (64, 49, 25, 24, (8, 64)), |
| (128, 3, 1, 1, (8, 128)), |
| (64, 3, 1, 1, (8, 64)), |
| ), |
| n_channel_embeddings=19, |
| n_global_tokens=1, |
| global_token_layer=1, |
| drop_prob=0.1, |
| ) |
|
|
|
|
