Merlin: A Vision Language Foundation Model for 3D Computed Tomography.

Over 85 million computed tomography (CT) scans are performed annually in the US, of which approximately one quarter focus on the abdomen. Given the current shortage of both general and specialized radiologists, there is a large impetus to use artificial intelligence to alleviate the burden of interpreting these complex imaging studies while simultaneously using the images to extract novel physiological insights. Prior state-of-the-art approaches for automated medical image interpretation leverage vision language models (VLMs) that utilize both the image and the corresponding textual radiology reports. However, current medical VLMs are generally limited to 2D images and short reports. To overcome these shortcomings for abdominal CT interpretation, we introduce Merlin - a 3D VLM that leverages both structured electronic health records (EHR) and unstructured radiology reports for pretraining without requiring additional manual annotations. We train Merlin using a high-quality clinical dataset of paired CT scans (6+ million images from 15,331 CTs), EHR diagnosis codes (1.8+ million codes), and radiology reports (6+ million tokens) for training. We comprehensively evaluate Merlin on 6 task types and 752 individual tasks. The non-adapted (off-the-shelf) tasks include zero-shot findings classification (31 findings), phenotype classification (692 phenotypes), and zero-shot cross-modal retrieval (image to findings and image to impressions), while model adapted tasks include 5-year chronic disease prediction (6 diseases), radiology report generation, and 3D semantic segmentation (20 organs). We perform internal validation on a test set of 5,137 CTs, and external validation on 7,000 clinical CTs and on two public CT datasets (VerSe, TotalSegmentator). Beyond these clinically-relevant evaluations, we assess the efficacy of various network architectures and training strategies to depict that Merlin has favorable performance to existing task-specific baselines. We derive data scaling laws to empirically assess training data needs for requisite downstream task performance. Furthermore, unlike conventional VLMs that require hundreds of GPUs for training, we perform all training on a single GPU. This computationally efficient design can help democratize foundation model training, especially for health systems with compute constraints. We plan to release our trained models, code, and dataset, pending manual removal of all protected health information.

Predicting cardiovascular disease risk using photoplethysmography and deep learning.

Cardiovascular diseases (CVDs) are responsible for a large proportion of premature deaths in low- and middle-income countries. Early CVD detection and intervention is critical in these populations, yet many existing CVD risk scores require a physical examination or lab measurements, which can be challenging in such health systems due to limited accessibility. We investigated the potential to use photoplethysmography (PPG), a sensing technology available on most smartphones that can potentially enable large-scale screening at low cost, for CVD risk prediction. We developed a deep learning PPG-based CVD risk score (DLS) to predict the probability of having major adverse cardiovascular events (MACE: non-fatal myocardial infarction, stroke, and cardiovascular death) within ten years, given only age, sex, smoking status and PPG as predictors. We compare the DLS with the office-based refit-WHO score, which adopts the shared predictors from WHO and Globorisk scores (age, sex, smoking status, height, weight and systolic blood pressure) but refitted on the UK Biobank (UKB) cohort. All models were trained on a development dataset (141,509 participants) and evaluated on a geographically separate test (54,856 participants) dataset, both from UKB. DLS's C-statistic (71.1%, 95% CI 69.9-72.4) is non-inferior to office-based refit-WHO score (70.9%, 95% CI 69.7-72.2; non-inferiority margin of 2.5%, p<0.01) in the test dataset. The calibration of the DLS is satisfactory, with a 1.8% mean absolute calibration error. Adding DLS features to the office-based score increases the C-statistic by 1.0% (95% CI 0.6-1.4). DLS predicts ten-year MACE risk comparable with the office-based refit-WHO score. Interpretability analyses suggest that the DLS-extracted features are related to PPG waveform morphology and are independent of heart rate. Our study provides a proof-of-concept and suggests the potential of a PPG-based approach strategies for community-based primary prevention in resource-limited regions.

Assistive AI in Lung Cancer Screening: A Retrospective Multinational Study in the United States and Japan.

Purpose To evaluate the impact of an artificial intelligence (AI) assistant for lung cancer screening on multinational clinical workflows. Materials and Methods An AI assistant for lung cancer screening was evaluated on two retrospective randomized multireader multicase studies where 627 (141 cancer-positive cases) low-dose chest CT cases were each read twice (with and without AI assistance) by experienced thoracic radiologists (six U.S.-based or six Japan-based radiologists), resulting in a total of 7524 interpretations. Positive cases were defined as those within 2 years before a pathology-confirmed lung cancer diagnosis. Negative cases were defined as those without any subsequent cancer diagnosis for at least 2 years and were enriched for a spectrum of diverse nodules. The studies measured the readers' level of suspicion (on a 0-100 scale), country-specific screening system scoring categories, and management recommendations. Evaluation metrics included the area under the receiver operating characteristic curve (AUC) for level of suspicion and sensitivity and specificity of recall recommendations. Results With AI assistance, the radiologists' AUC increased by 0.023 (0.70 to 0.72; P = .02) for the U.S. study and by 0.023 (0.93 to 0.96; P = .18) for the Japan study. Scoring system specificity for actionable findings increased 5.5% (57% to 63%; P < .001) for the U.S. study and 6.7% (23% to 30%; P < .001) for the Japan study. There was no evidence of a difference in corresponding sensitivity between unassisted and AI-assisted reads for the U.S. (67.3% to 67.5%; P = .88) and Japan (98% to 100%; P > .99) studies. Corresponding stand-alone AI AUC system performance was 0.75 (95% CI: 0.70, 0.81) and 0.88 (95% CI: 0.78, 0.97) for the U.S.- and Japan-based datasets, respectively. Conclusion The concurrent AI interface improved lung cancer screening specificity in both U.S.- and Japan-based reader studies, meriting further study in additional international screening environments. Keywords: Assistive Artificial Intelligence, Lung Cancer Screening, CT Supplemental material is available for this article. Published under a CC BY 4.0 license.

Total cholesterol in dyslipidemias. Is it a useful measure? / Colesterol total en dislipidemias. ¿Útil medirlo?

Abstract Introduction: the relationship between lipid fractions and cardiovascular risk is clear. However, the operational characteristics of total cholesterol (TC) for the diagnosis of dyslipidemias due to elevated LDL cholesterol (LDLC), hypertriglyceridemia and low HDL cholesterol (HDLC) are not clear. Objective: to establish the sensitivity (Sen) specificity (Spe) and predictive values (PPV and NPV) of TC (>200 mg/dL) for diagnosing various types of dyslipidemias. Materials and methods: a study of diagnostic tests using all the lipid profiles processed at the Hospital Universitario San Ignacio in Bogotá (Colombia) from January 2006 to January 2017. Sensitivity, Spe, PPV and NPV were calculated for each dyslipidemia and for each LDLC goal. Results: in 25,754 profiles, the average age was 53.6±18 years. The prevalence of elevated LDLC (based on the goals of 160, 130, 100, 70 or 55 mg/dL) was: 19.9%, 44.5%, 72.7%, 92.1% and 96.8%, respectively; for hypertriglyceridemia (>150 mg/dL) it was 44.7%, and for low HDLC (< 40 mg/dL) it was 33.9%. The sensitivity of TC (>200 mg/dL) for elevated LDLC according to the same goals was: 100%, 95%, 70%, 56% and 53%, with a specificity of: 59%, 81%, 94%, 95% and 92%; PPV=37%, 80%, 97%, 99% and 99%; and NPV=100%, 95%, 54%, 15% and 5.8%. For hypertrygliceridemia: Sen=61%, Spe=61%, PPV=55% and NPV=66%. For low HDLC: Sen=36%, Spe=42%, PPV=26% and NPV=54%. Conclusions: given the operational characteristics of TC>200 mg/dL, it should not be used as an isolated tool for diagnosing dyslipidemia due to LDLC, HDLC or hypertriglyceridemia. (Acta Med Colomb 2019; 44. DOI:https://doi.org/10.36104/amc.2019.1320).

Resumen Introducción: es clara la relation entre las fracciones lipídicas y riesgo cardiovascular, sin embargo, no son claras las características operativas del colesterol total (CT) para el diagnóstico de dislipidemias por colesterol LDL (C-LDL) elevado, hipertrigliceridemia y colesterol HDL (C-HDL) bajo. Objetivo: establecer sensibilidad (S), especificidad (E), y valores predictivos (VPP y VPN) del CT (>200 mg/dL) para diagnóstico de diferentes tipos de dislipidemias. Material y métodos: estudio de pruebas diagnosticas a partir de la totalidad de perfiles lipídicos procesados en el Hospital Universitario San Ignacio de Bogotá (Colombia), desde enero de 2006 hasta enero de 2017. Se calcularon S, E, VPP y VPN para cada dislipidemia y para cada meta de C-LDL. Resultados: en 25 754 perfiles, la edad promedio fue 53.6±18 años. Las prevalencias de C-LDL elevado (según metas de 160, 130, 100, 70 o 55 mg/dL) fueron: 19.9%, 44.5%, 72.7%, 92.1% y 96.8% respectivamente; hipertrigliceridemia (>150 mg/dL) 44.7% y C-HDL bajo (< 40 mg/dL) 33.9%. Las sensibilidades del CT (>200 mg/dL) para C-LDL elevado según las mismas metas fueron: 100%, 95%, 70%, 56% y 53% y especificidades: 59%, 81%, 94%, 95% y 92%. VPP=37%, 80%, 97%, 99% y 99%, y VPN=100%, 95%, 54%, 15% y 5.8%. Para hipertrigliceridemia: S=61%, E=61%, VPP=55% y VPN=66%. Para C-HDL bajo: S=36%, E=42%, VPP=26% y VPN=54%. Conclusiones: dadas las características operativas del CT>200 mg/dL, éste no debe ser utilizado como herramienta aislada para el diagnóstico de dislipidemia por C-LDL, por C-HDL, ni para hipertrigliceridemia. (Acta Med Colomb 2019; 44. DOI:https://doi.org/10.36104/amc.2019.1320).

Author Correction: End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography.

With an estimated 160,000 deaths in 2018, lung cancer is the most common cause of cancer death in the United States1. Lung cancer screening using low-dose computed tomography has been shown to reduce mortality by 20-43% and is now included in US screening guidelines1-6. Existing challenges include inter-grader variability and high false-positive and false-negative rates7-10. We propose a deep learning algorithm that uses a patient's current and prior computed tomography volumes to predict the risk of lung cancer. Our model achieves a state-of-the-art performance (94.4% area under the curve) on 6,716 National Lung Cancer Screening Trial cases, and performs similarly on an independent clinical validation set of 1,139 cases. We conducted two reader studies. When prior computed tomography imaging was not available, our model outperformed all six radiologists with absolute reductions of 11% in false positives and 5% in false negatives. Where prior computed tomography imaging was available, the model performance was on-par with the same radiologists. This creates an opportunity to optimize the screening process via computer assistance and automation. While the vast majority of patients remain unscreened, we show the potential for deep learning models to increase the accuracy, consistency and adoption of lung cancer screening worldwide.

Factores predictores perioperatorios de complicaciones de la colecistectomía por laparoscopia / Perioperative complications predictive risk factors in laparoscopic cholecystectomy

Introducción. Los factores de riesgo para la conversión a laparotomía y las complicaciones de la colecistectomía laparoscópica se han estudiado, pero no se conocen modelos actuales de predicción para estos resultados. Objetivo. Desarrollar un modelo de predicción para las complicaciones de la colecistectomía laparoscópica. Pacientes y métodos. Se llevó a cabo un estudio analítico retrospectivo que incluyó 1.234 pacientes con colelitiasis sometidos a colecistectomía laparoscópica, en un periodo de 18 meses en un hospital de IV nivel de Colombia. Se hizo el análisis multivariado por medio de regresión logística, usando el procedimiento backward para selección de variables, buscando determinar la probabilidad en un punto compuesto de complicación (presencia de, al menos, una complicación: lesión de vía biliar, colección intraabdominal o sangrado). Se elaboró una curva ROC para determinar la capacidad predictiva del modelo y el análisis de datos se hizo en Stata 13™. Resultados. Los pacientes incluidos se clasificaron en cohortes de derivación (926) y de validación (308), y se encontró que el 69,2 % eran mujeres; la edad mediana fue de 48 años (RIC=34-60) y, la conversión, de 4,3 %; hubo colección intraabdominal en 2,6 % y complicaciones en 4,7 %, y la mortalidad global fue de 0,3 %. La edad, la diabetes mellitus, la enfermedad renal crónica, la coledocolitiasis y el síndrome de Mirizzi se identificaron como predictores de alguna complicación. La capacidad predictiva del modelo fue de 58 %. Conclusión. La probabilidad de alguna complicación perioperatoria de la colecistectomía por laparoscopia depende de la edad, la diabetes mellitus, la enfermedad renal crónica, la coledocolitiasis y el síndrome de Mirizzi. Los factores de riesgo y las complicaciones descritos aquí, pueden guiar nuevas investigaciones y proporcionar evaluación de riesgos específicos

Introduction. Risk factors associated with conversion from laparoscopic cholecystectomy to open surgery and its complications are well known. However, there are currently no prediction models for such outcomes. The objective of tis work was to devolop a prediction model for laparoscopic cholecystectomy complications. Materials and methods. This is a retrospective analytical study that included 1,234 patients who underwent laparoscopic cholecystectomy in an 18 months period at a fourth level of care hospital in Bogota, Colombia. A multivariable logistic regression analysis using backward procedure was performed to for the selection of variables, in order to determine the likelihood of a combined endpoint complication (presence of at least one of the complications: bile ducts injury, haemorrhage, organ/space surgical site infection). A ROC curve was performed to determine the predictive ability of the model; information analysis was performed in 13 STATA. Results. Patients were classified in a derivation (926) and a validation cohort (308). It was found that 69.2 % were female, median age 48 years (IQR 34-60 ), conversion rate 4.3%, organ/space surgical site infections 2.6%, combined end point complication 4.7%, and global mortality rate 0.3%. Diabetes mellitus (DM), chronic kidney disease (CKD), choledocholitiasis and Mirizzi´s syndrome were found as predictors of the occurrence of complications. The model was validated in the validation cohort, obtaining an area under the ROC curve of 58%. Discussion. The likelihood of major complication in laparoscopic cholecystectomy depends on age, DM, CKD, choledocholitiasis, and Mirizzi´s syndrome. Risk Factors and complications described here can guide a new research avenue and provide the evaluation of specific risks

Síndrome de Wellens: reconociendo el peligro / Wellens' syndrome: recognizing the risks

El infarto agudo de miocardio es la principal causa de muerte en el mundo. Las guías de manejo de esta patología indican el estudio invasivo urgente en los casos de infarto con supradesnivel del ST o bloqueo nuevo de rama izquierda. Se han descrito patrones electrocardiográficos, que cuando se presentan en el contexto clínico de angina inestable o infarto, predicen la presencia de obstrucciones coronarias severas que pueden ocasionar la muerte. El reconocimiento de estos patrones es clave para su manejo. Se describe el síndrome de Wellens como parte de estas alteraciones predictoras de riesgo.

Acute myocardial infarction is the leading cause of death worldwide. Dedicated management guidelines indicate urgent invasive study in cases of infarction with ST elevation or new left bundle branch block. Electrocardiographic patterns when present in the clinical setting of unstable angina or myocardial infarction predict the presence of severe coronary disease that can lead to death. The recognition of these patterns is very important to take management decisions. A description of Wellens' syndrome as part of these high-risk predictors is offered.

Deep neural networks rival the representation of primate IT cortex for core visual object recognition.

The primate visual system achieves remarkable visual object recognition performance even in brief presentations, and under changes to object exemplar, geometric transformations, and background variation (a.k.a. core visual object recognition). This remarkable performance is mediated by the representation formed in inferior temporal (IT) cortex. In parallel, recent advances in machine learning have led to ever higher performing models of object recognition using artificial deep neural networks (DNNs). It remains unclear, however, whether the representational performance of DNNs rivals that of the brain. To accurately produce such a comparison, a major difficulty has been a unifying metric that accounts for experimental limitations, such as the amount of noise, the number of neural recording sites, and the number of trials, and computational limitations, such as the complexity of the decoding classifier and the number of classifier training examples. In this work, we perform a direct comparison that corrects for these experimental limitations and computational considerations. As part of our methodology, we propose an extension of "kernel analysis" that measures the generalization accuracy as a function of representational complexity. Our evaluations show that, unlike previous bio-inspired models, the latest DNNs rival the representational performance of IT cortex on this visual object recognition task. Furthermore, we show that models that perform well on measures of representational performance also perform well on measures of representational similarity to IT, and on measures of predicting individual IT multi-unit responses. Whether these DNNs rely on computational mechanisms similar to the primate visual system is yet to be determined, but, unlike all previous bio-inspired models, that possibility cannot be ruled out merely on representational performance grounds.