Pathology detection

Name: pathology

Category: clinical

Dataset: Lopez2017 (TUAB)

Objective: Multiclass classification

Split: Predefined

Usage

neuralbench eeg pathology

Show config.yaml

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.

data:
  study:
    source:
      name: Lopez2017Tuab
    crop_timelines:
      name: CropTimelines
      event_type: Eeg
      start_offset_s: 60.0
      max_duration_s: 1200.0
    split:
      name: PredefinedSplit
      test_split_query: "split == 'eval'"
      col_name: split
      valid_split_by: subject
      valid_split_ratio: 0.2
      valid_random_state: 33
  neuro:
    picks:
      - Fp1
      - Fp2
      - F7
      - F8
      - F3
      - Fz
      - F4
      - T9
      - T7
      - C3
      - Cz
      - C4
      - T8
      - T10
      - P7
      - P3
      - Pz
      - P4
      - P8
      - O1
      - O2
  target:
    =replace=: true
    name: LabelEncoder
    event_types: Eeg
    event_field: label
    return_one_hot: true
  trigger_event_type: Eeg
  start: 0.0
  duration: 5.0
  stride: 5.0
  summary_columns: [label]
compute_class_weights: true
brain_model_output_size: &brain_model_output_size 2
trainer_config.monitor: val/bal_acc
trainer_config.mode: max
loss:
  name: CrossEntropyLoss
  kwargs:
    label_smoothing: 0.1
metrics: !!python/object/apply:neuralbench.defaults.metrics.get_classification_metric_configs
  - *brain_model_output_size
test_full_metrics:
  - log_name: acc
    name: Accuracy
    kwargs:
      task: multiclass
      num_classes: *brain_model_output_size
  - log_name: bal_acc
    name: Accuracy
    kwargs:
      task: multiclass
      num_classes: *brain_model_output_size
      average: macro
  - log_name: auroc
    name: AUROC
    kwargs:
      task: multiclass
      num_classes: *brain_model_output_size

Description

The goal of the pathology detection task is to predict whether an EEG recording contains pathological EEG data or not [Lopez2017].

Dataset Notes

• Previous work (BIOT, LaBraM, CBraMod, etc.) evaluates test performance at the window level, rather than at the recording level like domain-specific SOTA papers (e.g. [Gemein2020]). Here, we evaluate performance at the recording level.

References

[Lopez2017]

Lopez, Sebas, et al. “Automated identification of abnormal adult EEGs.” 2015 IEEE signal processing in medicine and biology symposium (SPMB). IEEE, 2015.

[Gemein2020]

Gemein, Lukas AW, et al. “Machine-learning-based diagnostics of EEG pathology.” NeuroImage 220 (2020): 117021.