Chin Siang Ong

Background: Quantitative assessment of aortic morphology from computed tomography (CT) underpins diagnosis, surveillance, and risk stratification in aortic disease. However, existing analysis workflows are often fragmented, algorithm-specific, and limited to discrete vessel segments, constraining reproducibility, longitudinal comparison, and integration of emerging computational methods. The objective of this study was to describe AortaPlot, a modular, end-to-end computational framework designed to standardize continuous aortic quantification and generate anatomically contextualized outputs suitable for clinical interpretation, as well as downstream machine learning and longitudinal modeling. Methods: AortaPlot ingests CT imaging data and performs three-dimensional aortic segmentation using a replaceable segmentation engine. A continuous vascular centerline is computed from the aortic root to the aortic bifurcation and parameterized by arc length. Orthogonal cross-sectional planes are resampled at configurable intervals, and vessel diameter is quantified using a geometry-adaptive Sliding Axis algorithm that identifies the maximum orthogonal diameter independent of cross-sectional shape. Twenty anatomical landmarks are automatically detected and embedded within the measurement framework. The system generates landmark-annotated diameter-versus-length profiles, high-resolution regional analyses, standardized radiomic feature sets, annotated full length aortic visualization and fixed-view multi-angle 3D surface visualizations. Results: Across representative normal and pathological cases, AortaPlot reproducibly reconstructed full-length centerlines and generated anatomically ordered, landmark-annotated diameter profiles spanning the entire aorta. Continuous profiling captured physiologic tapering as well as focal aneurysmal dilation, while region-specific oversampling enabled detailed characterization of localized pathology. Outputs are compiled into standardized, surgeon-oriented reports integrating continuous quantitative profiles, focused segment analysis, anatomically contextualized radiomic features, landmark measurements, annotated aortic visualization and six-view 3D visualizations with fixed rotational spacing. Conclusions: AortaPlot establishes a reproducible, extensible framework for continuous aortic morphology analysis that unifies segmentation, measurement, anatomical context, and visualization within a single pipeline. By standardizing representation across the full vessel length and across time, this protocol reduces analytic variability and produces structured outputs directly amenable to radiomics, machine learning, and longitudinal modeling, providing a scalable foundation for reproducible aortic surveillance and future AI-driven risk stratification.

Background: Clinical decision support requires language models that provide guideline-aligned, context-aware reasoning with clear justification. Many existing benchmarks emphasize multiple-choice or short-form question answering and mainly capture factual recall rather than longitudinal clinical reasoning from extended clinical notes. Hippocrates-o1 is a family of domain-tailored clinical reasoning pipelines that combine structured prompts, guideline-informed retrieval, and iterative self-refinement across oncology, general surgery, and vascular surgery. Methods: Real-world head and neck cancer cases were drawn from the MIMIC-IV-Note database, with a subset (n=20) randomly selected for detailed annotation. Six physicians adjudicated treatment phase and intent using structured criteria and rated model outputs. For each case, we generated outputs using both a general-purpose baseline model ( VanillaLLM ) and our oncology-specific reasoning model, Hippocrates-Karkinos-o1 . Experts evaluated the outputs across five dimensions on a scale of 1 to 5: Clinical Knowledge Application, Contextual Understanding, Reasoning Transparency, Chain-of-Thought Quality, and Hallucination Audit. Overall Reasoning was the mean of domain scores. To explore whether the approach could extend beyond oncology, we also processed inguinal hernia and aortic aneurysm cases through Hippocrates-Chirurgos-o1 and Hippocrates-Angios-o1 domain adaptations. Results: Across paired ratings, Hippocrates-Karkinos-o1 improved Overall Reasoning from 3.40±0.90 to 4.00±0.73 (p<0.001). Domain scores increased for Clinical Knowledge Application (2.87±1.20 to 3.70±1.03), Contextual Understanding (3.48±0.95 to 3.98±0.95), Hallucination Audit (3.90±1.32 to 4.74±0.76), Reasoning Transparency (3.45±1.02 to 3.86±0.87), and Chain-of-Thought Quality (3.32±1.04 to 3.69±1.00), all p≤0.001. Surgical and vascular adaptations showed parallel qualitative improvements. Conclusions: The Hippocrates-o1 protocol improved reasoning fidelity, guideline alignment, and factual grounding relative to a general-purpose model and generalized across oncology, surgery, and vascular care. Orchestrated retrieval and self-refinement provide a reproducible template for evaluating and enhancing clinical reasoning in medical AI.

Chest computed tomography (CT) is commonly used to evaluate thoracic aortic dilations and aneurysms, but aortic measurements are often embedded in unstructured radiology report narratives, limiting their accessibility for clinical decision-making and research. Structured extraction of these measurements could flag incidental findings in the electronic health record, improve adherence to surveillance and surgical referral guidelines, and support population-level studies on aortic disease. We compared the performance of fine-tuned BERT-based models with instruction-tuned, few-shot, and zero-shot Llama large language models for extracting aortic diameters from 356,690 institutional chest CT reports spanning 2013 to 2023. A subset of 2010 reports was manually annotated, with 1002 used for training, 504 for validation, and 504 for testing. Aortic diameters were annotated at eight standardized anatomical sites. The instruction-tuned Llama 3.1 model achieved the highest performance, with a macro F1 score of 0.992 on the validation set and 0.970 on the test set, outperforming the few-shot (F1 = 0.838) and zero-shot (F1 = 0.663) Llama 3.1 models, as well as the fine-tuned Clinical BERT model (F1 = 0.954). Applied to the full dataset, the instruction-tuned model identified aortic measurements in 49,387 of 356,690 reports (13.85%). These results demonstrate that instruction-tuned large language models enable high-fidelity extraction of clinically relevant information with minimal annotation, and the framework developed here is adaptable to a broad range of medical information extraction tasks from unstructured text.

Because the lung is a mucosal barrier organ with a unique immunologic environment, mechanisms of immunoregulation in lung cancer may differ from those of other malignancies. Consistent with this notion, we found that CD8 ⁺ T cells played a paradoxical role in facilitating, rather than ameliorating, the growth of multiple lung adenocarcinoma models. These included spontaneous, carcinogen-induced, and transplantable tumor cell line models. Specifically, we found that CD8 ⁺ T cells promoted homing of CD4 ⁺ Foxp3 ⁺ Tregs to the tumor bed by increasing the levels of CCR5 chemokines in the tumor microenvironment in an IFN-γ– and TNF-α–dependent manner. Contrary to their canonical role, these Th1 cytokines contributed to accelerated growth of murine lung adenocarcinomas, while suppressing the growth of other malignancies. Surprisingly, lung cancer cells themselves can serve as a dominant source of IFN-γ, and deletion of this cytokine from cancer cells using CRISPR/Cas9 decreases tumor growth. Importantly for translational applications, in patients with lung cancer, a high level of IFN-γ was also found at both the mRNA and protein levels. Our data outline what we deem a novel and previously undefined lung cancer–specific immunoregulatory pathway that may be harnessed to tailor immune-based therapy specifically for this malignancy.

Objectives The rising global burden of cardiovascular diseases (CV) highlights the critical need for efficiency in disease diagnosis and management. An important area for such improvement is utilization of artificial intelligence (AI) for streamlining time and resources in CV imaging workflow. We evaluate the performance of artificial intelligence (AI) segmentation for aortic segmentation on clinical computed tomography angiography (CTA) images and compare accuracy to manual methods. Such automation would markedly improve efficiency and accuracy of aortic surveillance. Methods This retrospective study included 27 scans from 20 patients who underwent thoracic endovascular aortic repair (TEVAR) between January 2020 and March 2022. An open source AI model was applied to segment the aorta, and its performance was assessed by comparing AI-generated segmentations with manual segmentations using Dice similarity coefficients, volumetric analysis, and aortic dimensions. Centerline reconstructed images of thoracoabdominal aorta were processed to extract radiomic features, including maximum diameter and cross-sectional area, for analysis. Results The AI tool achieved a mean Dice coefficient of 0.96 ± 0.03, indicating a high degree of concordance with manual segmentation. Multiplanar reconstruction was performed to visualize the aorta and extract measurements along its length using the automated centerline, and radiomic features, including maximum diameter and cross-sectional area, were subsequently extracted for analysis. Conclusions AI segmentation demonstrates strong potential for improving efficiency and consistency in thoracoabdominal aortic segmentation, achieving high accuracy compared to manual methods. These findings highlight the feasibility of AI integration into clinical practice for diagnosis and surveillance of aortopathies, warranting further validation on larger datasets to enable clinical translation.

Background: Risk stratification models estimate the probabilities of adverse outcomes after cardiac surgical procedures, which helps clinicians and patients make informed decisions. Objectives: We used the 12-lead electrocardiogram (ECG) and/or Society for Thoracic Surgeons (STS) variables to predict postoperative outcomes using deep learning methods that can incorporate diverse data types. Methods: We developed a deep convolutional neural network (“ECGNet”) that predicts operative mortality and other adverse outcomes using preoperative 12-lead ECGs (n=30,877) from 12,933 patients who underwent 13,299 cardiac surgical procedures. We also developed a deep neural network applied to preoperative STS variables (“STSNet”). STSNet and ECGNet are multi-task neural networks that utilize secondary outcomes to augment prediction of mortality using the same neural network. Results: ECGNet achieved a mean area under the receiver operating characteristic curve (AUC) of 0.85 for predicting operative mortality for all procedures and 0.93 for valve procedures. STSNet achieved a mean AUC of 0.85 for all procedures, with statistically similar performance as ECGNet for all procedures. Combining ECGNet and STSNet achieved a mean AUC of 0.90 for predicting operative mortality after all procedures, which is significantly higher than either ECGNet or STSNet alone. Conclusions: A deep neural network trained on STS features has higher predictive performance than previously reported for existing conventional models and is not limited to certain types of cardiac surgical procedures. A model trained on ECG alone can predict operative mortality with similar performance as STS features and adding ECG to STS features in a neural network can improve performance. These findings demonstrate the potential in leveraging deep learning on multidimensional data sources to predict outcomes after cardiac surgery. Condensed In this study, deep learning (DL) is applied to electrocardiograms and clinical features used in the standard STS risk prediction tools to generate new high-performing risk calculators for cardiac surgical procedures. Preoperative voltage waveforms contain information about cardiovascular risk and cardiac function and are passed as inputs to the deep learning model. These risk models apply to all cardiac procedures including those procedures that do not have standard STS risk calculators and provide improved performance. DL models enable the incorporation of additional modalities of data to improve risk prediction in cardiac surgery.

Chest computed tomography (CT) is essential for diagnosing and monitoring thoracic aortic dilations and aneurysms, conditions that place patients at risk of complications such as aortic dissection and rupture. However, aortic measurements in chest CT radiology reports are often embedded in free-text formats, limiting their accessibility for clinical care, quality improvement and research purposes. In this study, we developed a multi-method pipeline to extract structured aortic measurements from radiology reports, and compared the performance of fine-tuned BERT-based models with instruction-tuned Llama large language models (LLMs). Applying the best-performing method to a real-world large chest CT radiology report database, we generated a comprehensive aortic measurement dataset that facilitates big data aortic disease research.

SurgeryLLM, a large language model framework using Retrieval Augmented Generation demonstrably incorporated domain-specific knowledge from current evidence-based surgical guidelines when presented with patient-specific data. The successful incorporation of guideline-based information represents a substantial step toward enabling greater surgeon efficiency, improving patient safety, and optimizing surgical outcomes.

SurgeryLLM, a large language model framework using Retrieval Augmented Generation demonstrably incorporated domain-specific knowledge from current evidence-based surgical guidelines when presented with patient-specific data. The successful incorporation of guideline-based information represents a substantial step toward enabling greater surgeon efficiency, improving patient safety, and optimizing surgical outcomes.