Learning, Potential, and Retention: An Approach for Evaluating Adaptive AI-Enabled Medical Devices

Alexis Burgon; Berkman Sahiner; Nicholas A Petrick; Gene Pennello; Ravi K Samala

cs.AIApr 6, 2026

Learning, Potential, and Retention: An Approach for Evaluating Adaptive AI-Enabled Medical Devices

Authors: Alexis Burgon, Berkman Sahiner, Nicholas A Petrick, Gene Pennello, Ravi K Samala

Organizations: Division of Imaging, Diagnostics, and Software Reliability (DIDSR), Office of Science and Engineering Laboratories (OSEL), Center for Devices and Radiological Health (CDRH), U.S. Food and Drug Administration, 10903 New Hampshire Ave, Silver Spring, MD US 20993

Abstract

This work addresses challenges in evaluating adaptive artificial intelligence (AI) models for medical devices, where iterative updates to both models and evaluation datasets complicate performance assessment. We introduce a novel approach with three complementary measurements: learning (model improvement on current data), potential (dataset-driven performance shifts), and retention (knowledge preservation across modification steps), to disentangle performance changes caused by model adaptations versus dynamic environments. Case studies using simulated population shifts demonstrate the approach's utility: gradual transitions enable stable learning and retention, while rapid shifts reveal trade-offs between plasticity and stability. These measurements provide practical insights for regulatory science, enabling rigorous assessment of the safety and effectiveness of adaptive AI systems over sequential modifications.

Figures & tables

Figure 1: A simple adaptation timeline for an adaptive AI system with an initial implementation (timepoint 0) and five modification steps (timepoints 1-5), showing the transition between timepoints 2 (current) and 3 (new).

Figure 2: Comparison of learning between Scenario A & Scenario B using a toy example. Despite showing the same performance at each comparable modification step (0 & 1), as indicated by their identical line plots, the two scenarios exhibit different learning . The performance change in Scenario A is due to a shift in dataset difficulty, whereas the performance change in Scenario B is due to an improvement in model knowledge.

Figure 3: Comparison of potential between Scenario A & Scenario B using a toy example. Despite showing the same performance at each comparable modification step (0 & 1), as indicated by their identical line plots, the two scenarios exhibit different potential . Scenario A demonstrates greater potential because the modification step 1 dataset presented a greater challenge to the modification step 0 model than was observed in Scenario B.

Figure 4: Comparison of retention between Scenario A & Scenario B using a toy example. Despite showing the same performance at comparable modification step (0 & 1), as indicated by their identical line plots, the two scenarios exhibit different retention . Scenario A shows lower retention because the modification step 1 model demonstrates greater performance degradation on the modification step 0 evaluation dataset.

	Training	Validation	Testing
Disease Positive	525	75	150
Disease Negative	525	75	150
Total	1050	150	300

Table 1: Number of patients included in each data partition (training, validation, testing) for each modification step. Images from a single patient were not split across different partitions.

Panel of Figure 5 — Figure 5: (a) Population distribution of training, validation, and testing data, (b) learning & potential and (c) retention & performance for a model trained and evaluated on a dataset gradually transitioning from one population to another. Vertical markers indicate 95% confidence intervals from models across 25 repetitions.

Panel of Figure 6 — Figure 6: (a) Population distribution of training, validation, and testing data, (b) learning & potential and (c) retention & performance for a model with limited plasticity trained and evaluated on a dataset gradually transitioning from one population to another. Vertical markers indicate 95% confidence intervals from models across 25 repetitions.

Panel of Figure 7 — Figure 7: (a) Population distribution of training, validation, and testing data, (b) learning & potential and (c) retention & performance for a model trained and evaluated on a dataset gradually transitioning between three populations. Vertical markers indicate 95% confidence intervals from models across 25 repetitions.

Learning	Measures the effect of model modification with respect to the current evaluation dataset.
	High learning means that the modifications improved model performance for the current population, but it may also indicate a lack of consistency in the model over time. If previous datasets still represent a portion of the current intended population, retention may need to be measured.
	Negative learning indicates that the current model performs worse than the previous model on the current dataset. If this is observed alongside an increase in performance, the new dataset may merit examination.
Potential	Indicates the challenge posed by the new dataset from the perspective of the prior model.
	High potential may indicate the addition of a new population. If a new population was not deliberately added, a population shift may have occurred.
Retention	Indicates the model’s performance on previous datasets.
	Low retention may be a concern when previous datasets represent a portion of the current intended population.

Table 2: Overview of the interpretation and potential utilities of the proposed learning , potential , and retention measurements.

Explore similar work

Robust by Design: A Continuous Monitoring and Data Integration Framework for Medical AI
Apr 10, 2026Mohammad Daouk, Jan Ulrich Becker, Neeraja Kambham +3Online Continual Learning Incremental Learning
University of Houston, Houston, TX, USA · University Hospital Cologne, Cologne, Germany · Stanford University, Stanford, CA, USA +1
AEGIS: An Operational Infrastructure for Post-Market Governance of Adaptive Medical AI Under US and EU Regulations
Mar 20, 2026Fardin Afdideh, Mehdi Astaraki, Fernando Seoane +1AI Governance Responsible AI
Department of Clinical Science, Intervention and Technology, Karolinska Institutet, 17177 Stockholm, Sweden · Department of Medical Radiation Physics, Stockholm University, Solna, Sweden · Department of Oncology-Pathology, Karolinska Institutet, Solna, Sweden +4
Audit Before You Update: Recording Artifacts and the Limits of Bounded Model Updates in MIMIC-IV ICU Prediction
Jul 21, 2026Fatema Ferdous Tamanna, K. M. Merajul Arefin, Md. Abdul MasudClinical Decision Support Longitudinal EHR Modeling
Dept. of Computer Science and IT, Patuakhali Science and Technology University, Bangladesh · Dept. of Computer Science and Engineering, University of Dhaka, Bangladesh

Apr 10, 2026J/K move · Enter open · S save

Robust by Design: A Continuous Monitoring and Data Integration Framework for Medical AI

Mohammad Daouk, Jan Ulrich Becker, Neeraja Kambham, Anthony Chang+2

University of Houston, Houston, TX, USA · University Hospital Cologne, Cologne, Germany · Stanford University, Stanford, CA, USA+1

Online Continual Learning Incremental Learning Medical Image Classification

Open paper AskPDF arXiv

Adaptive medical AI models often face performance drops in dynamic clinical environments due to data drift. We propose an autonomous continuous monitoring and data integration framework that maintains robust performance over time. Focusing on glomerular pathology image classification (proliferative vs. non-proliferative lupus nephritis), our three-stage method uses multi-metric feature analysis and Monte Carlo dropout-based uncertainty gating to decide when to retrain on new data. Only images statistically similar to the training distribution (via Euclidean, cosine, Mahalanobis metrics) and with low predictive entropy are integrated. The model is then incrementally retrained with these images under strict performance safeguards (no metric degradation >5%). In experiments with a ResNet18 ensemble on a multi-center dataset, the framework prevents performance degradation: new images were added without significant change in AUC (~0.92) or accuracy (~89%). This approach addresses data shift and avoids catastrophic forgetting, enabling sustained learning in medical imaging AI.

Abstract

Explore similar work

Robust by Design: A Continuous Monitoring and Data Integration Framework for Medical AI

AEGIS: An Operational Infrastructure for Post-Market Governance of Adaptive Medical AI Under US and EU Regulations

Audit Before You Update: Recording Artifacts and the Limits of Bounded Model Updates in MIMIC-IV ICU Prediction