Biology subjects

Hammer, H. L.

Publications and source records attributed to Hammer, H. L..

5 recordsLinked to original sources

bioRxiv2025

Learning Residual-based Biomarkers of Cognitive Health via Self-Supervised Learning on EEG State Transitions

Deep learning (DL) models have achieved impressive performance in EEG-based prediction tasks, but they often lack interpretability, limiting their clinical utility. In this study, we introduce a novel self-supervised learning (SSL) framework inspired by neurophysiological reactivity. Our approach models healthy EEG transitions between ocular states by predicting an EEG-derived feature under eyes-open conditions, using features from eyes-closed recordings. The residual between observed and predicted values quantifies deviations from normative brain dynamics and serves as a candidate biomarker. To improve clinical relevance, we propose two optimisation strategies that promote residuals predictive of pathology. We evaluated the framework using healthy cohorts (LEMON, Dortmund Vital) for self-supervised training, and the AI-Mind cohort for downstream prediction of plasma p-tau217 levels, a proxy for cognitive pathology. Despite extensive hyperparameter optimisation, predictive performance remained poor across all methods, including baseline models, suggesting limitations in the downstream proxy or input signal. Nonetheless, our approach provides a transparent methodology for transition-based EEG biomarker discovery grounded in self-supervised learning.

neuroscience↗

bioRxiv2025

Uncertainty in Deep Learning for EEG under Dataset Shifts

ObjectiveAs artificial intelligence (AI) is increasingly integrated into medical diagnostics, it is essential that predictive models provide not only accurate outputs but also reliable estimates of uncertainty. In clinical applications, where decisions have significant consequences, understanding the confidence behind each prediction is as critical as the prediction itself. Uncertainty modelling plays a key role in improving trust, guiding decision-making, and identifying unreliable outputs, particularly under dataset shift or in out-of-distribution settings. The primary aim of uncertainty metrics is to align model confidence closely with actual predictive performance, ensuring confidence estimates dynamically adjust to reflect increasing errors or decreasing reliability of predictions. This study investigates how different ensemble learning strategies affect both performance and uncertainty estimation in a clinically relevant task: classifying Normal, Mild Cognitive Impairment, and Dementia from electroencephalography (EEG) data. ApproachWe evaluated the performance and uncertainty of ensemble methods and Monte Carlo dropout on a large EEG dataset. The models were assessed in three settings: (1) in-distribution performance on a held-out test set, (2) generalisation to three out-of-distribution datasets, and (3) performance under gradual, EEG-specific dataset shifts simulating noise, drift, and frequency perturbation. Main resultsEnsembles consisting of multiple independently trained models, such as deep ensembles, consistently achieved higher performance in both the in-distribution test set and the out-of-distribution datasets. These models also produced more informative and responsive uncertainty estimates under various types of EEG dataset shifts. SignificanceThese results highlight the benefits of ensemble diversity and independent training to build robust and uncertainty-aware EEG classification models. The findings are particularly relevant for clinical applications, where reliability under distribution shift and transparent uncertainty are essential for safe deployment.

neuroscience↗

bioRxiv2025

Assessing the robustness of deep learning based brain age prediction models across multiple EEG datasets

The increasing availability of large electroencephalography (EEG) datasets enhances the potential clinical utility of deep learning (DL) for cognitive and pathological decoding. However, dataset shifts due to variations in the population and acquisition hardware can considerably degrade the model performance. We systematically investigated the generalisation of DL models to unseen datasets with different characteristics, using age as the target variable. Five datasets were used in two different experimental setups, including (1) leave-one-dataset-out (LODO) and (2) leave-one-dataset-in (LODI) cross validation. A comprehensive set of 1805 different hyperparameter configurations was tested, including variations in the DL architectures and data pre-processing. The performance varied across source/target dataset pair. Using LODO, we obtained Pearsons r values of {0.63, 0.84, 0.75, 0.23, 0.10} and R2 values of {-0.01, 0.63, 0.41, -4.66, -70.98}. For LODI, the results varied in Pearsons r from -0.11 to 0.84 and R2 values from -704.89 to 0.65, depending on the source and target dataset. Adjusting the model intercepts using the average age of the target dataset substantially improved some R2 scores. Our results show that DL models can learn age-related EEG patterns which generalise with strong correlations to datasets with broad age spans. The most important hyperparameter was to use the frequency range between 1 and 45Hz, rather than a single frequency band. The second most important hyperparameter effect depended on the experimental setup. Our findings highlight the challenges of dataset shifts in EEG-based DL models and establish a benchmark for future studies aiming to improve the robustness of DL models across diverse datasets.

neuroscience↗

bioRxiv2024

EEG Electrodes and Where to Find Them: Automated Localization From 3D Scans

ObjectiveThe accurate localization of electroencephalography (EEG) electrode positions is crucial for accurate source localization. Recent advancements have proposed alternatives to labor-intensive, manual methods for spatial localization of the electrodes, employing technologies such as 3D scanning and laser scanning. These novel approaches often integrate Magnetic Resonance Imaging (MRI) as part of the pipeline in localizing the electrodes. The limited global availability of MRI data restricts its use as a standard modality in several clinical scenarios. This limitation restricts the use of these advanced methods. ApproachIn this paper, we present a novel, versatile approach that utilizes 3D scans to localize EEG electrode positions with high accuracy. Importantly, while our method can be integrated with MRI data if available, it is specifically designed to be highly effective even in the absence of MRI, thus expanding the potential for advanced EEG analysis in various resource-limited settings. Our solution implements a two-tiered approach involving landmark/fiducials localization and electrode localization, creating an end-to-end framework. Main resultsThe efficacy and robustness of our approach have been validated on an extensive dataset containing over 400 3D scans from 278 subjects. The framework identifies pre-auricular points and achieves correct electrode positioning accuracy in the range of 85.7% to 91.0%. Additionally, our framework includes a validation tool that permits manual adjustments and visual validation if required. SignificanceThis study represents, to the best of the authors knowledge, the first validation of such a method on a substantial dataset, thus ensuring the robustness and generalizability of our innovative approach. Our findings focus on developing a solution that facilitates source localization, contributing to the critical discussion on balancing cost effectiveness with methodological accuracy to promote wider adoption in both research and clinical settings.

neuroscience↗

bioRxiv2024

Deep learning methods to forecasting human embryo development in time-lapse videos

BackgroundIn assisted reproductive technology, evaluating the quality of the embryo is crucial when selecting the most viable embryo for transferring to a woman. Assessment also plays an important role in determining the optimal transfer time, either in the cleavage stage or in the blastocyst stage. Several AI-based tools exist to automate the assessment process. However, none of the existing tools predicts upcoming video frames to assist embryologists in the early assessment of embryos. In this paper, we propose an AI system to forecast the dynamics of embryo morphology over a time period in the future. MethodsThe AI system is designed to analyze embryo development in the past two hours and predict the morphological changes of the embryo for the next two hours. It utilizes a predictive model incorporating Convolutional LSTM layers, to predict the future video frame by analyzing prior morphological changes within the embryos video sequence. The system uses the predictions recursively and forecasts up to 23 hours of embryo development. ResultsThe results demonstrated that the AI system could accurately forecast embryo development at the cleavage stage on day 2 and the blastocyst stage on day 4. The system provided valuable information on the cell division processes on day 2 and the start of the blastocyst stage on day 4. The system focused on specific developmental features effective across both the categories of embryos. The embryos that were transferred to the female, and the embryos that were discarded. However, in the transfer category, the forecast had a clearer cell membrane and less distortion as compared to the avoid category. ConclusionThis study assists in the embryo evaluation process by providing early insights into the quality of the embryo for both the transfer and avoid categories of videos. The embryologists recognize the ability of the forecast to depict the morphological changes of the embryo. Additionally, enhancement in image quality has the potential to make this approach relevant in clinical settings. Author summaryThe emergence of assisted reproductive technology has significantly improved infertility treatments. It involves fertilization of an egg outside the body, and the resultant embryos are developed in time-lapse incubators. The embryologists manually evaluate embryos using time-lapse videos and rank each embryo on the basis of several criteria including the dynamics of embryo cell stages, such as the start of the blastocyst stage. Traditional manual analysis is subjective and time-consuming, and AI tools are introduced to automate and enhance embryo selection efficiency. However, current AI tools do not generate video frames that forecast changes in embryo morphology. This study fills this gap by introducing an AI system that forecasts upcoming frames of a time-lapse video. In this approach, several hours were predicted ahead of the last video frame. The system was evaluated on crucial days of embryo evaluation. Our approach was effective in both good quality (transfer) and poor quality (avoid) video categories, and the forecast revealed crucial insights about embryo cell division and the start of the blastocyst stage. Despite some image quality issues, the proposed AI system demonstrated the potential for early and accurate assessment of embryo quality.

cell biology↗