skip to main content


This content will become publicly available on April 3, 2025

Title: Quantifying input data drift in medical machine learning models by detecting change-points in time-series data
Devices enabled by artificial intelligence (AI) and machine learning (ML) are being introduced for clinical use at an accelerating pace. In a dynamic clinical environment, these devices may encounter conditions different from those they were developed for. The statistical data mismatch between training/initial testing and production is often referred to as data drift. Detecting and quantifying data drift is significant for ensuring that AI model performs as expected in clinical environments. A drift detector signals when a corrective action is needed if the performance changes. In this study, we investigate how a change in the performance of an AI model due to data drift can be detected and quantified using a cumulative sum (CUSUM) control chart. To study the properties of CUSUM, we first simulate different scenarios that change the performance of an AI model. We simulate a sudden change in the mean of the performance metric at a change-point (change day) in time. The task is to quickly detect the change while providing few false-alarms before the change-point, which may be caused by the statistical variation of the performance metric over time. Subsequently, we simulate data drift by denoising the Emory Breast Imaging Dataset (EMBED) after a pre-defined change-point. We detect the change-point by studying the pre- and post-change specificity of a mammographic CAD algorithm. Our results indicate that with the appropriate choice of parameters, CUSUM is able to quickly detect relatively small drifts with a small number of false-positive alarms.  more » « less
Award ID(s):
2326034
NSF-PAR ID:
10539737
Author(s) / Creator(s):
; ; ; ; ; ; ;
Editor(s):
Astley, Susan M; Chen, Weijie
Publisher / Repository:
SPIE
Date Published:
ISBN:
9781510671584
Page Range / eLocation ID:
11
Format(s):
Medium: X
Location:
San Diego, United States
Sponsoring Org:
National Science Foundation
More Like this
  1. Connected vehicle (CV) systems are cognizant of potential cyber attacks because of increasing connectivity between its different components such as vehicles, roadside infrastructure and traffic management centers. However, it is a challenge to detect security threats in real-time and develop appropriate/effective countermeasures for a CV system because of the dynamic behavior of such attacks, high computational power requirement and a historical data requirement for training detection models. To address these challenges, statistical models, especially change point models, have potentials for real-time anomaly detections. Thus, the objective of this study is to investigate the efficacy of two change point models, Expectation Maximization (EM) and two forms of Cumulative Summation (CUSUM) algorithms (i.e., typical and adaptive), for real-time V2I cyber attack detection in a CV Environment. To prove the efficacy of these models, we evaluated these two models for three different type of cyber attack, denial of service (DOS), impersonation, and false information, using basic safety messages (BSMs) generated from CVs through simulation. Results from numerical analysis revealed that EM, CUSUM, and adaptive CUSUM could detect these cyber attacks, DOS, impersonation, and false information, with an accuracy of (99\%, 100\%, 100\%), (98\%, 100\%, 100\%), and (100\%, 98\%, 100\%) respectively. 
    more » « less
  2. The cumulative sum (CUSUM) control chart is a method for detecting whether the mean of a time series process has shifted beyond some tolerance (ie, is out of control). Originally developed in an industrial process control setting, the CUSUM statistic is typically reset to zero once a process is discovered to be out of control since the industrial process is then recalibrated to be in control. The CUSUM method is also used to detect disease outbreaks in prospective disease surveillance, with a disease outbreak coinciding with an out‐of‐control process. In a disease surveillance setting, resetting the CUSUM statistic is unrealistic, and a nonrestarting CUSUM chart is used instead. In practice, the nonrestarting CUSUM provides more information but suffers from a high false alarm rate following the end of an outbreak. In this paper, we propose a modified hypothesis test for use with the nonrestarting CUSUM when testing whether a process is out of control. By simulating statistics conditional on the presence of an out‐of‐control process in recent time periods, we are able to retain the CUSUM's power to detect an out‐of‐control process while controlling the post–out‐of‐control false alarm rate at the desired level. We demonstrate this method using data on aSalmonellaNewport outbreak that occurred in Germany in 2011. We find that in 7 out of 8 states where the outbreak was detected, the outbreak was detected at the same speed as an unmodified nonrestarting CUSUM while controlling the postoutbreak rate of false alarms at the desired level.

     
    more » « less
  3. False power consumption data injected from compromised smart meters in Advanced Metering Infrastructure (AMI) of smart grids is a threat that negatively affects both customers and utilities. In particular, organized and stealthy adversaries can launch various types of data falsification attacks from multiple meters using smart or persistent strategies. In this paper, we propose a real time, two tier attack detection scheme to detect orchestrated data falsification under a sophisticated threat model in decentralized micro-grids. The first detection tier monitors whether the Harmonic to Arithmetic Mean Ratio of aggregated daily power consumption data is outside a normal range known as safe margin. To confirm whether discrepancies in the first detection tier is indeed an attack, the second detection tier monitors the sum of the residuals (difference) between the proposed ratio metric and the safe margin over a frame of multiple days. If the sum of residuals is beyond a standard limit range, the presence of a data falsification attack is confirmed. Both the ‘safe margins’ and the ‘standard limits’ are designed through a ‘system identification phase’, where the signature of proposed metrics under normal conditions are studied using real AMI micro-grid data sets from two different countries over multiple years. Subsequently, we show how the proposed metrics trigger unique signatures under various attacks which aids in attack reconstruction and also limit the impact of persistent attacks. Unlike metrics such as CUSUM or EWMA, the stability of the proposed metrics under normal conditions allows successful real time detection of various stealthy attacks with ultra-low false alarms. 
    more » « less
  4. High false alarm rate in intensive care units (ICUs) has been identified as one of the most critical medical challenges in recent years. This often results in overwhelming the clinical staff by numerous false or unurgent alarms and decreasing the quality of care through enhancing the probability of missing true alarms as well as causing delirium, stress, sleep deprivation and depressed immune systems for patients. One major cause of false alarms in clinical practice is that the collected signals from different devices are processed individually to trigger an alarm, while there exists a considerable chance that the signal collected from one device is corrupted by noise or motion artifacts. In this paper, we propose a low-computational complexity yet accurate game-theoretic feature selection method which is based on a genetic algorithm that identifies the most informative biomarkers across the signals collected from various monitoring devices and can considerably reduce the rate of false alarms. 
    more » « less
  5. Sequential change detection is a classical problem with a variety of applications. However, the majority of prior work has been parametric, for example, focusing on exponential families. We develop a fundamentally new and general framework for sequential change detection when the pre- and post-change distributions are nonparametrically specified (and thus composite). Our procedures come with clean, nonasymptotic bounds on the average run length (frequency of false alarms). In certain nonparametric cases (like sub-Gaussian or sub-exponential), we also provide near-optimal bounds on the detection delay following a changepoint. The primary technical tool that we introduce is called an e-detector, which is composed of sums of e-processes—a fundamental generalization of nonnegative supermartingales—that are started at consecutive times. We first introduce simple Shiryaev-Roberts and CUSUM-style e-detectors, and then show how to design their mixtures in order to achieve both statistical and computational efficiency. Our e-detector framework can be instantiated to recover classical likelihood-based procedures for parametric problems, as well as yielding the first change detection method for many nonparametric problems. As a running example, we tackle the problem of detecting changes in the mean of a bounded random variable without i.i.d. assumptions, with an application to tracking the performance of a basketball team over multiple seasons.

     
    more » « less