arXiv subjects

Peter G. Jacobs

Publications and source records attributed to Peter G. Jacobs.

6 recordsLinked to original sources

arXiv2026

Explainable AI-based Decision Support for Nocturnal Hypoglycemia Prevention in Type 1 Diabetes

Purpose: Nocturnal hypoglycemia (NH) remains a challenge for individuals with type 1 diabetes (T1D), particularly those who are physically active or on multiple daily injections (MDI). We leveraged an explainable evidential neural network model that forecasts minimum overnight glucose to identify NH risk factors and generate recommendations for NH prevention. Methods: We analyzed the impact of glucose and physical activity (PA) factors on predicted NH probability using SHapley Additive exPlanations (SHAP). Data were from 20 adults with T1D on MDI (10 females; mean age 39 years; HbA1c 7\%) who participated in a cross-over study (NCT05967260). Results: A total of 502 nights were analyzed. Bedtime glucose was the strongest predictor of NH. Other relevant factors included recent exposure to high or low glucose, glucose variability before bedtime, and timing of PA. Some associations appeared physiologically counterintuitive, possibly reflecting behavioral adjustments. Based on the identified risk factors and their SHAP values, we determined key decision points and developed recommendations that may help prevent NH, such as consuming a bedtime snack or discussing potential adjustments to insulin therapy with a healthcare provider. Conclusion: Identifying predictors of NH offers insights for managing NH risk in clinical decision support.

q-bio.QM↗

arXiv2026

PhysioSeq2Seq: A Hybrid Physiological Digital Twin and Sequence-to-Sequence LSTM for Long-Horizon Glucose Forecasting in Type 1 Diabetes

Accurate long-horizon glucose forecasting is critical for automated insulin delivery systems, which help people with type 1 diabetes (T1D) manage their glucose and avoid dangerous hypoglycemia. However, standard recursive long short-term memory (LSTM) networks suffer from systematic negative bias at longer horizons due to error compounding, while purely mechanistic ordinary differential equation (ODE) models fail to generalize across individuals when parameterized at the population level. We propose PhysioSeq2Seq, a hybrid architecture that combines patient-specific physiological modeling with a sequence-to-sequence (Seq2Seq) LSTM. For each glucose segment, twin matching searches a population of 300 parameterized digital twins to identify the best-fitting physiological match from a 3-hour continuous glucose monitoring (CGM) history. The 10 internal ODE state variables of the matched twin are injected as exogenous covariates into both the encoder and decoder of the Seq2Seq LSTM. This simultaneous 48-step prediction strategy eliminates recursive error compounding, while the ODE features provide a physics-grounded constraint that bounds long-horizon drift within physiologically plausible ranges. PhysioSeq2Seq was trained on CGM and insulin data from 348 participants in the Type 1 Diabetes Exercise Initiative (T1DEXI) dataset and evaluated on 74 held-out participants. At the 240-minute horizon, PhysioSeq2Seq achieves a mean absolute error of 39.28 mg/dL and a mean error of -10.62 mg/dL, reducing bias by 13.89 mg/dL over the recursive LSTM and reducing mean absolute error by 28.62 mg/dL over the ODE-based digital twin. These results show that eliminating architectural feedback and injecting patient-matched physiological states is an effective and clinically meaningful strategy for long-horizon glucose forecasting in T1D.

cs.LG↗

arXiv2025

A Physiologically-Constrained Neural Network Digital Twin Framework for Replicating Glucose Dynamics in Type 1 Diabetes

Simulating glucose dynamics in individuals with type 1 diabetes (T1D) is critical for developing personalized treatments and supporting data-driven clinical decisions. Existing models often miss key physiological aspects and are difficult to individualize. Here, we introduce physiologically-constrained neural network (NN) digital twins to simulate glucose dynamics in T1D. To ensure interpretability and physiological consistency, we first build a population-level NN state-space model aligned with a set of ordinary differential equations (ODEs) describing glucose regulation. This model is formally verified to conform to known T1D dynamics. Digital twins are then created by augmenting the population model with individual-specific models, which include personal data, such as glucose management and contextual information, capturing both inter- and intra-individual variability. We validate our approach using real-world data from the T1D Exercise Initiative study. Two weeks of data per participant were split into 5-hour sequences and simulated glucose profiles were compared to observed ones. Clinically relevant outcomes were used to assess similarity via paired equivalence t-tests with predefined clinical equivalence margins. Across 394 digital twins, glucose outcomes were equivalent between simulated and observed data: time in range (70-180 mg/dL) was 75.1$\pm$21.2% (simulated) vs. 74.4$\pm$15.4% (real; P<0.001); time below range (<70 mg/dL) 2.5$\pm$5.2% vs. 3.0$\pm$3.3% (P=0.022); and time above range (>180 mg/dL) 22.4$\pm$22.0% vs. 22.6$\pm$15.9% (P<0.001). Our framework can incorporate unmodeled factors like sleep and activity while preserving key dynamics. This approach enables personalized in silico testing of treatments, supports insulin optimization, and integrates physics-based and data-driven modeling. Code: https://github.com/mosqueralopez/T1DSim_AI

q-bio.QM↗

arXiv2025

An AI-enabled dual-hormone model predictive control algorithm that delivers insulin and pramlintide

Current closed-loop insulin delivery algorithms need to be informed of carbohydrate intake disturbances. This can be a burden on people using these systems. Pramlintide is a hormone that delays gastric emptying, which enables insulin kinetics to align with the kinetics of carbohydrate absorption. Integrating pramlintide into an automated insulin delivery system can be helpful in reducing the postprandial glucose excursion and may be helpful in enabling fully-closed loop whereby meals do not need to be announced. We present an AI-enabled dual-hormone model predictive control (MPC) algorithm that delivers insulin and pramlintide without requiring meal announcements that uses a neural network to automatically detect and deliver meal insulin. The MPC algorithm includes a new pramlintide pharmacokinetics and pharmacodynamics model that was identified using data collected from people with type 1 diabetes undergoing a meal challenge. Using a simulator, we evaluated the performance of various pramlintide delivery methods and controller models, as well as the baseline insulin-only scenario. Meals were automatically dosed using a neural network meal detection and dosing (MDD) algorithm. The primary outcome was the percent time glucose is in the target range (TIR: 70-180 mg/dL). Results show that delivering pramlintide at a fixed ratio of 6 mcg pramlintide:1 u insulin using an MPC that is aware of the pramlintide achieved the most significant improved TIR compared with an insulin-only MPC (91.6% vs. 64.1%). Delivering pramlintide as a fixed ratio was better than delivering basal pramlintide at a constant rate, indicating the benefit of the MDD algorithm. There was no advantage of independent control of insulin and pramlintide compared with insulin and pramlintide delivered as a fixed ratio. Preliminary real-world results from a human subject further indicate the benefit of pramlintide delivery.

eess.SY↗

arXiv2015

Gait Velocity Estimation using time interleaved between Consecutive Passive IR Sensor Activations

Gait velocity has been consistently shown to be an important indicator and predictor of health status, especially in older adults. It is often assessed clinically, but the assessments occur infrequently and do not allow optimal detection of key health changes when they occur. In this paper, we show that the time gap between activations of a pair of Passive Infrared (PIR) motion sensors installed in the consecutively visited room pair carry rich latent information about a person's gait velocity. We name this time gap transition time and show that despite a six second refractory period of the PIR sensors, transition time can be used to obtain an accurate representation of gait velocity. Using a Support Vector Regression (SVR) approach to model the relationship between transition time and gait velocity, we show that gait velocity can be estimated with an average error less than 2.5 cm/sec. This is demonstrated with data collected over a 5 year period from 74 older adults monitored in their own homes. This method is simple and cost effective and has advantages over competing approaches such as: obtaining 20 to 100x more gait velocity measurements per day and offering the fusion of location-specific information with time stamped gait estimates. These advantages allow stable estimates of gait parameters (maximum or average speed, variability) at shorter time scales than current approaches. This also provides a pervasive in-home method for context-aware gait velocity sensing that allows for monitoring of gait trajectories in space and time.

cs.OH↗

arXiv2014

Fabrication and Characterization of an Amperometric Glucose Sensor on a Flexible Polyimide Substrate

This study details the use of printing and other additive processes to fabricate a novel amperometric glucose sensor. The sensor was fabricated using a Au coated 12.7 micron polyimide film as a starting material, where micro-contact printing, electrochemical plating and chloridization, electrohydrodynamic jet (e-jet) printing, and spin coating were used to pattern, deposit, print, and coat functional materials, respectively. We have found that e-jet printing was effective for the deposition and patterning of glucose oxidase inks between ~5 to 1000 micron in width, and we have demonstrated that the enzyme was still active after printing. The thickness of the permselective layer was optimized to obtain a linear response to glucose concentration up to 32 mM. For these sensors no response to acetaminophen, a common interfering compound, was observed.

cond-mat.mtrl-sci↗