| { |
| "meta_info": { |
| "case_id": "ISRUC_02.edf", |
| "bench_subset": "NeuroBench-Sleep", |
| "difficult": 2, |
| "original_dataset": "ISRUC" |
| }, |
| "agent_input": { |
| "data_path": "data/sleep/ISRUC_02.edf", |
| "label_path": "data/sleep/ISRUC_02.npy", |
| "instruction": "Given the sleep EEG/PSG file and paired sleep-stage label file, assess whole-night respiratory event clustering. Divide the recording into fixed, non-overlapping 5-minute windows starting at 0 minutes: 0-5, 5-10, 10-15, and so on. Use only complete 5-minute windows. Count apnea and hypopnea candidate events by their event onset time; an event belongs to exactly one half-open window [start, end). A cluster window is a 5-minute window with at least 3 apnea/hypopnea candidate events. For each cluster window, determine the dominant sleep stage from the ten 30-second sleep-stage labels in that window. If the dominant stage is tied, use the sleep stages at the candidate event onset epochs within the window to break the tie; if still tied, use W, N1, N2, N3, R order. Report the total number of cluster windows and the dominant stage across all cluster windows. If no cluster windows occur, report dominant stage as NULL. Allowed stage outputs are W, N1, N2, N3, R, and NULL. Respiratory channels for ISRUC: X5 = snore; X6 = Flow-1 (pressure-based airflow); DC3 = Flow-2 airflow; X7 and X8 = abdominal respiratory effort." |
| }, |
| "eval_config": { |
| "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for sleep medicine reports.\n\n### TASK\nExtract the total respiratory cluster-window count and dominant cluster sleep stage.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. Do not include Markdown, explanations, units, or extra fields.\n3. The keys must be EXACTLY \"cluster_window_count\" and \"dominant_cluster_stage\".\n4. \"cluster_window_count\" must be a FLOAT or null.\n5. \"dominant_cluster_stage\" must be one of \"W\", \"N1\", \"N2\", \"N3\", \"R\", \"NULL\", or null.\n6. If there are no respiratory cluster windows, normalize the stage to \"NULL\".\n7. If a value is missing or unclear, return null for that key.\n\n### OUTPUT TEMPLATE\n{\"cluster_window_count\": <float|null>, \"dominant_cluster_stage\": <\"W\"|\"N1\"|\"N2\"|\"N3\"|\"R\"|\"NULL\"|null>}", |
| "metrics": [ |
| { |
| "metric_id": "respiratory_cluster_window_count_accuracy", |
| "type": "numeric_check", |
| "target_key": "cluster_window_count", |
| "weight": 60, |
| "params": { |
| "gt_value": 3.0, |
| "tolerance": 0.0 |
| } |
| }, |
| { |
| "metric_id": "respiratory_cluster_dominant_stage_accuracy", |
| "type": "categorical_check", |
| "target_key": "dominant_cluster_stage", |
| "weight": 40, |
| "params": { |
| "gt_value": "N1" |
| } |
| } |
| ] |
| } |
| } |
