The BHI2026 DATA Challenge Competition (Track 2) aims to promote the development of generalizable AI algorithms for ambulatory blood pressure (BP) and circadian pattern estimation. Through a standardized, high-quality dataset and a reproducible competition framework, the challenge seeks to establish a transparent benchmark, encourage methodological innovation, and help bridge the gap between algorithm development and real-world wearable cardiovascular monitoring.
Submissions are evaluated using a composite metric called Overall_MAE, which comprises three complementary components of blood pressure estimation performance: Segment_MAE, Period_MAE, and Dip_MAE. The formula is defined as:
Overall_MAE = 0.4 · Segment_MAE + 0.3 · Period_MAE + 0.3 · Dip_MAE
Segment_MAE (Segment-level Accuracy): Calculated across all test segments. For each segment, it is defined as:
Segment_MAE = 0.6 · |SBP_true − SBP_est| + 0.4 · |DBP_true − DBP_est|
Period_MAE (Day/Night Period Accuracy): Calculated across all subjects as the average of Day and Night accuracy:
Period_MAE = 0.5 · Day_MAE + 0.5 · Night_MAE
Where daytime is defined as 08:00–20:00 and nighttime as 22:00–08:00. Both Day_MAE and Night_MAE follow the same weighting structure as Segment_MAE but use average values for the respective periods.
Dip_MAE (Nocturnal Dip Accuracy): Also calculated across all subjects, measuring the accuracy of the nocturnal dip in blood pressure:
Dip_MAE = 0.6 · |DipSBP_true − DipSBP_est| + 0.4 · |DipDBP_true − DipDBP_est|
Where Dip_SBP and Dip_DBP represent the absolute difference between day and night average systolic and diastolic pressures, respectively.
Lower overall MAE indicates better model performance. For leaderboard ranking, the Overall_MAE is transformed into a score ranging from 0 to 1 using the formula Leaderboard Score = 1 / (1 + Overall_MAE). While the Public Leaderboard provides interim feedback, final rankings are determined solely by the Private Leaderboard.