OBJECTIVES: Leveraging routine electronic health records (EHR) for dementia detection is a growing field, but quality and clinical utility of existing models are unclear. This systematic review aimed to evaluate performance, methodological quality, and risk of bias of EHR-based dementia prediction models. MATERIALS AND METHODS: We systematically searched Medline, EMBASE, Scopus, IEEE Xplore, and ACM from conception until July 2024. All studies and grey literature describing development or validation of probabilistic prediction models using EHR data for dementia detection were included. Risk of bias was assessed using PROBAST. RESULTS: Fifty-six studies (434 prediction models, 155 external validations) were included. Most models were prognostic (66%), used US data (71%), relied solely on structured data, and 47 (11%) were externally validated. Modeled outcomes were extremely heterogeneous: gold-standard clinical criteria were used in 17 models (4%), with others reliant on diagnostic codes for case ascertainment. Discriminative metrics were frequently reported (82% of models), but calibration was rarely assessed (16%). All models were judged high risk of bias, driven by poor outcome definition, inadequate handling of missing data, and potential overfitting. DISCUSSION: Our review highlights significant issues with methodological rigor and reporting transparency in existing EHR dementia prediction models. Ambiguous outcomes, flawed case ascertainment, and incomplete performance reporting, all limit clinical usefulness. Overall, model performance was difficult to assess and compare across studies due to incomplete reporting. CONCLUSION: Electronic health record-based dementia prediction is still in its infancy. Methodological rigor and interdisciplinary collaboration are essential to meet clinical needs and achieve real-world impact.

Current work in named entity recognition (NER) uses either cross entropy (CE) or conditional random fields (CRF) as the objective/loss functions to optimize the underlying NER model. Both of these traditional objective functions for the NER problem generally produce adequate performance when the data distribution is balanced and there are sufficient annotated training examples. But since NER is inherently an imbalanced tagging problem, the model performance under the low-resource settings could suffer using these standard objective functions. Based on recent advances in area under the ROC curve (AUC) maximization, we propose to optimize the NER model by maximizing the AUC score. We give evidence that by simply combining two binary-classifiers that maximize the AUC score, significant performance improvement over traditional loss functions is achieved under low-resource NER settings. We also conduct extensive experiments to demonstrate the advantages of our method under the low-resource and highly-imbalanced data distribution settings. To the best of our knowledge, this is the first work that brings AUC maximization to the NER setting. Furthermore, we show that our method is agnostic to different types of NER embeddings, models and domains. The code of this work is available at https://github.com/dngu0061/NER-AUC-2T.

Named entity recognition (NER), a task that identifies and categorizes named entities such as persons or organizations from text, is traditionally framed as a multi-class classification problem. However, this approach often overlooks the issues of imbalanced label distributions, particularly in low-resource settings, which is common in certain NER contexts, like biomedical NER (bioNER). To address these issues, we propose an innovative reformulation of the multi-class problem as a one-vs-all (OVA) learning problem and introduce a loss function based on the area under the receiver operating characteristic curve (AUC). To enhance the efficiency of our OVA-based approach, we propose two training strategies: one groups labels with similar linguistic characteristics, and another employs meta-learning. The superiority of our approach is confirmed by its performance, which surpasses traditional NER learning in varying NER settings.

Domain adaptation is an effective solution to data scarcity in low-resource scenarios.However, when applied to token-level tasks such as bioNER, domain adaptation methods often suffer from the challenging linguistic characteristics that clinical narratives possess, which leads to unsatsifactory performance.In this paper, we present a simple yet effective hardnessguided domain adaptation (HGDA) framework for bioNER tasks that can effectively leverage the domain hardness information to improve the adaptability of the learnt model in the low-resource scenarios.Experimental results on biomedical datasets show that our model can achieve significant performance improvement over the recently published state-of-theart (SOTA) MetaNER model.

The watershed transform, from mathematical morphology, is a powerful and flexible tool for segmentation. However, it does not allow a priori knowledge relating to characteristics of region boundaries to be included in the way that other approaches do. This paper introduces the locally constrained watershed transform, which includes border constraints by modifying the underlying path definition upon which the watershed transform depends. This approach maintains many of the desirable properties of the watershed transform, such as well-defined stopping conditions and efficient implementation, while offering more stable segmentation in the presence of noisy or incomplete boundaries.

The mechanisms employed to detect motion by a variety of biological systems have been investigated for many years and a number of models that explain different aspects of these systems have been developed. Most of the models have been developed to explain aspects of wide field operation of motion detector arrays. This paper investigates suitability of these model as a front end to a system that requires local motion information. Results of simulations of different motion detection models using real scenes captured using a video camera are also presented.