Gabriel Davis Jones

Lin Li, Georgia Channing, Suhaas M Bhat et al.

Large language models (LLMs) have achieved remarkable progress in open-ended text generation, yet they remain prone to hallucinating incorrect or unsupported content, which undermines their reliability. This issue is exacerbated in long-form generation due to hallucination snowballing, a phenomenon where early errors propagate and compound into subsequent outputs. To address this challenge, we propose a novel inference-time hallucination mitigation framework, named Segment-wise HAllucination Rejection Sampling (SHARS), which uses an arbitrary hallucination detector to identify and reject hallucinated segments during generation and resample until faithful content is produced. By retaining only confident information and building subsequent generations upon it, the framework mitigates hallucination accumulation and enhances factual consistency. To instantiate this framework, we adopt semantic uncertainty as the detector and introduce several vital modifications to address its limitations and better adapt it to long-form text. Our method enables models to self-correct hallucinations without requiring external resources such as web search or knowledge bases, while remaining compatible with them for future extensions. Empirical evaluations on standardized hallucination benchmarks demonstrate that our method substantially reduces hallucinations in long-form generation while preserving or even improving the informativeness of generation. Code is available at: https://github.com/TreeLLi/hallucination-rejection-sampling.

M. Jaleed Khan, Manu Vatish, Gabriel Davis Jones

Antepartum Cardiotocography (CTG) is vital for fetal health monitoring, but traditional methods like the Dawes-Redman system are often limited by high inter-observer variability, leading to inconsistent interpretations and potential misdiagnoses. This paper introduces PatchCTG, a transformer-based model specifically designed for CTG analysis, employing patch-based tokenisation, instance normalisation and channel-independent processing to capture essential local and global temporal dependencies within CTG signals. PatchCTG was evaluated on the Oxford Maternity (OXMAT) dataset, comprising over 20,000 CTG traces across diverse clinical outcomes after applying the inclusion and exclusion criteria. With extensive hyperparameter optimisation, PatchCTG achieved an AUC of 77%, with specificity of 88% and sensitivity of 57% at Youden's index threshold, demonstrating adaptability to various clinical needs. Testing across varying temporal thresholds showed robust predictive performance, particularly with finetuning on data closer to delivery, achieving a sensitivity of 52% and specificity of 88% for near-delivery cases. These findings suggest the potential of PatchCTG to enhance clinical decision-making in antepartum care by providing a reliable, objective tool for fetal health assessment. The source code is available at https://github.com/jaleedkhan/PatchCTG.

Isabel Tsintsiper, Sheng Wong, Beth Albert et al.

Large language models (LLMs) are increasingly embedded in healthcare workflows for documentation, education, and clinical decision support. However, these systems are trained on large text corpora that encode existing biases, including sex disparities in diagnosis and treatment, raising concerns that such patterns may be reproduced or amplified. We systematically examined whether contemporary LLMs exhibit sex-specific biases in clinical reasoning and how model configuration influences these behaviours. We conducted three experiments using 50 clinician-authored vignettes spanning 44 specialties in which sex was non-informative to the initial diagnostic pathway. Four general-purpose LLMs (ChatGPT (gpt-4o-mini), Claude 3.7 Sonnet, Gemini 2.0 Flash and DeepSeekchat). All models demonstrated significant sex-assignment skew, with predicted sex differing by model. At temperature 0.5, ChatGPT assigned female sex in 70% of cases (95% CI 0.66-0.75), DeepSeek in 61% (0.57-0.65) and Claude in 59% (0.55-0.63), whereas Gemini showed a male skew, assigning a female sex in 36% of cases (0.32-0.41). Contemporary LLMs exhibit stable, model-specific sex biases in clinical reasoning. Permitting abstention reduces explicit labelling but does not eliminate downstream diagnostic differences. Safe clinical integration requires conservative and documented configuration, specialty-level clinical data auditing, and continued human oversight when deploying general-purpose models in healthcare settings.

M. Jaleed Khan, Ioana Duta, Beth Albert et al.

The rapid advancement of Artificial Intelligence (AI) in healthcare presents a unique opportunity for advancements in obstetric care, particularly through the analysis of cardiotocography (CTG) for fetal monitoring. However, the effectiveness of such technologies depends upon the availability of large, high-quality datasets that are suitable for machine learning. This paper introduces the Oxford Maternity (OxMat) dataset, the world's largest curated dataset of CTGs, featuring raw time series CTG data and extensive clinical data for both mothers and babies, which is ideally placed for machine learning. The OxMat dataset addresses the critical gap in women's health data by providing over 177,211 unique CTG recordings from 51,036 pregnancies, carefully curated and reviewed since 1991. The dataset also comprises over 200 antepartum, intrapartum and postpartum clinical variables, ensuring near-complete data for crucial outcomes such as stillbirth and acidaemia. While this dataset also covers the intrapartum stage, around 94% of the constituent CTGS are antepartum. This allows for a unique focus on the underserved antepartum period, in which early detection of at-risk fetuses can significantly improve health outcomes. Our comprehensive review of existing datasets reveals the limitations of current datasets: primarily, their lack of sufficient volume, detailed clinical data and antepartum data. The OxMat dataset lays a foundation for future AI-driven prenatal care, offering a robust resource for developing and testing algorithms aimed at improving maternal and fetal health outcomes.

Jahan C. Penny-Dimri, Magdalena Bachmann, William R. Cooke et al.

Large language models (LLMs) hold substantial promise for clinical decision support. However, their widespread adoption in medicine, particularly in healthcare, is hindered by their propensity to generate false or misleading outputs, known as hallucinations. In high-stakes domains such as women's health (obstetrics & gynaecology), where errors in clinical reasoning can have profound consequences for maternal and neonatal outcomes, ensuring the reliability of AI-generated responses is critical. Traditional methods for quantifying uncertainty, such as perplexity, fail to capture meaning-level inconsistencies that lead to misinformation. Here, we evaluate semantic entropy (SE), a novel uncertainty metric that assesses meaning-level variation, to detect hallucinations in AI-generated medical content. Using a clinically validated dataset derived from UK RCOG MRCOG examinations, we compared SE with perplexity in identifying uncertain responses. SE demonstrated superior performance, achieving an AUROC of 0.76 (95% CI: 0.75-0.78), compared to 0.62 (0.60-0.65) for perplexity. Clinical expert validation further confirmed its effectiveness, with SE achieving near-perfect uncertainty discrimination (AUROC: 0.97). While semantic clustering was successful in only 30% of cases, SE remains a valuable tool for improving AI safety in women's health. These findings suggest that SE could enable more reliable AI integration into clinical practice, particularly in resource-limited settings where LLMs could augment care. This study highlights the potential of SE as a key safeguard in the responsible deployment of AI-driven tools in women's health, leading to safer and more effective digital health interventions.

Sheng Wong, Ravi Shankar, Beth Albert et al.

Supervised deep learning models for automated CTG analysis are typically constrained by narrowly curated labelled datasets and limited patient cohorts, leaving substantial volumes of physiologically informative clinical recordings untapped. To address this limitation, we propose Physiology-aware Representation Learning via Integrated Self-supervision and Metadata for CTG (PRISM-CTG), a clinically grounded self-supervised foundation model (FM) for CTG that leverages large-scale unlabelled recordings to learn transferable domain-level representations. PRISM-CTG is pretrained using a multi-view self-supervised framework that jointly optimises 3 complementary pretext objectives: random-projected guided masked signal reconstruction, clinical variable prediction, and feature classification. Each objective is associated with a dedicated task-specific token, enabling specialised representation learning, while controlled cross-attention facilitates information exchange across clinical context. By reframing patient metadata and domain knowledge, which are often underutilised in conventional training as prediction targets, Prism-CTG transforms readily available clinical information into additional supervisory targets that guide clinically meaningful representation learning. Extensive experiments across 7 downstream CTG tasks in both antepartum and intrapartum domains demonstrated that PRISM-CTG consistently outperforms in-domain and SSL baselines. Notably, PRISM-CTG demonstrated strong generalisation under external validation on 2 datasets, while achieving comparable performance to studies trained on substantially larger, privately labelled datasets. To our knowledge, this is the first study to introduce large-scale FM for CTG that learns domain-level representations.

Sheng Wong, Beth Albert, Gabriel Davis Jones

Cardiotocography (CTG) is essential for fetal monitoring but is frequently compromised by diverse artefacts which obscure true fetal heart rate (FHR) patterns and can lead to misdiagnosis or delayed intervention. Current deep-learning approaches typically bypass comprehensive noise handling, applying minimal preprocessing or focusing solely on downstream classification, while traditional methods rely on simple interpolation or rule-based filtering that addresses only missing samples and fail to correct complex artefact types. We present CleanCTG, an end-to-end dual-stage model that first identifies multiple artefact types via multi-scale convolution and context-aware cross-attention, then reconstructs corrupted segments through artefact-specific correction branches. Training utilised over 800,000 minutes of physiologically realistic, synthetically corrupted CTGs derived from expert-verified "clean" recordings. On synthetic data, CleanCTG achieved perfect artefact detection (AU-ROC = 1.00) and reduced mean squared error (MSE) on corrupted segments to 2.74 x 10^-4 (clean-segment MSE = 2.40 x 10^-6), outperforming the next best method by more than 60%. External validation on 10,190 minutes of clinician-annotated segments yielded AU-ROC = 0.95 (sensitivity = 83.44%, specificity 94.22%), surpassing six comparator classifiers. Finally, when integrated with the Dawes-Redman system on 933 clinical CTG recordings, denoised traces increased specificity (from 80.70% to 82.70%) and shortened median time to decision by 33%. These findings suggest that explicit artefact removal and signal reconstruction can both maintain diagnostic accuracy and enable shorter monitoring sessions, offering a practical route to more reliable CTG interpretation.

Sheng Wong, Ravi Shankar, Beth Albert et al.

Foundation models (FMs) and large language models (LLMs) have demonstrated promising generalization across diverse domains for time-series analysis, yet their potential for electronic fetal monitoring (EFM) and cardiotocography (CTG) analysis remains underexplored. Most existing CTG studies relied on domain-specific models and lack systematic comparisons with modern foundation or language models, limiting our understanding of whether these models can outperform specialized systems in fetal health assessment. In this study, we present the first comprehensive benchmark of state-of-the-art architectures for automated antepartum CTG classification. Over 2,500 20-minutes recordings were used to evaluate over 15 models spanning domain-specific, time-series, foundation, and language-model categories under a unified framework. Fine-tuned LLMs consistently outperformed both foundation and domain-specific models across data-availability scenarios, except when uterine-activity signals were absent, where domain-specific models showed greater robustness. These performance gains, however, required substantially higher computational resources. Our results highlight that while fine-tuned LLMs achieved state-of-the-art performance for CTG classification, practical deployment must balance performance with computational efficiency.

John Tolladay, Beth Albert, Gabriel Davis Jones

Objective: To develop and interpret a supervised variational autoencoder (VAE) model for classifying cardiotocography (CTG) signals based on pregnancy outcomes, addressing interpretability limits of current deep learning approaches. Methods: The OxMat CTG dataset was used to train a VAE on five-minute fetal heart rate (FHR) segments, labeled with postnatal outcomes. The model was optimised for signal reconstruction and outcome prediction, incorporating Kullback-Leibler divergence and total correlation (TC) constraints to structure the latent space. Performance was evaluated using area under the receiver operating characteristic curve (AUROC) and mean squared error (MSE). Interpretability was assessed using coefficient of determination, latent traversals and unsupervised component analyses. Results: The model achieved an AUROC of 0.752 at the segment level and 0.779 at the CTG level, where predicted scores were aggregated. Relaxing TC constraints improved both reconstruction and classification. Latent analysis showed that baseline-related features (e.g., FHR baseline, baseline shift) were well represented and aligned with model scores, while metrics like short- and long-term variability were less strongly encoded. Traversals revealed clear signal changes for baseline features, while other properties were entangled or subtle. Unsupervised decompositions corroborated these patterns. Findings: This work demonstrates that supervised VAEs can achieve competitive fetal outcome prediction while partially encoding clinically meaningful CTG features. The irregular, multi-timescale nature of FHR signals poses challenges for disentangling physiological components, distinguishing CTG from more periodic signals such as ECG. Although full interpretability was not achieved, the model supports clinically useful outcome prediction and provides a basis for future interpretable, generative models.

Ravi Shankar, Sheng Wong, Lin Li et al.

Reliable abstention is critical for retrieval-augmented generation (RAG) systems, particularly in safety-critical domains such as women's health, where incorrect answers can lead to harm. We present an energy-based model (EBM) that learns a smooth energy landscape over a dense semantic corpus of 2.6M guideline-derived questions, enabling the system to decide when to generate or abstain. We benchmark the EBM against a calibrated softmax baseline and a k-nearest neighbour (kNN) density heuristic across both easy and hard abstention splits, where hard cases are semantically challenging near-distribution queries. The EBM achieves superior abstention performance abstention on semantically hard cases, reaching AUROC 0.961 versus 0.950 for softmax, while also reducing FPR@95 (0.235 vs 0.331). On easy negatives, performance is comparable across methods, but the EBM's advantage becomes most pronounced in safety-critical hard distributions. A comprehensive ablation with controlled negative sampling and fair data exposure shows that robustness stems primarily from the energy scoring head, while the inclusion or exclusion of specific negative types (hard, easy, mixed) sharpens decision boundaries but is not essential for generalisation to hard cases. These results demonstrate that energy-based abstention scoring offers a more reliable confidence signal than probability-based softmax confidence, providing a scalable and interpretable foundation for safe RAG systems.