Tanmay Gokhale

Aims: AI in electrocardiography (ECG) has diverged into two paths: traditional signal processing with machine learning, and deep learning of raw waveforms. The former preserves physiological interpretability but may miss novel patterns, while the latter overlooks a century of ECG science. No framework yet integrates both. This study introduces a fusion model combining handcrafted feature-based random forest and waveform-based convolutional neural network to improve the diagnosis of acute coronary events. Methods and results: This prospective, multicentre cohort study included pre-hospital ECGs recorded on emergency medical services from patients with a chief complaint of chest pain. Using this dataset, we developed four fusion frameworks that integrate ECG features with waveform information: (1) decision-level fusion combining predictions from random forest and convolutional neural network (CNN); (2) decision-level fusion with retraining; (3) feature-level fusion by concatenating CNN embeddings with ECG features; and (4) feature-level fusion combining CNN embeddings with tree-level outputs from the random forest. This study comprised of 10 393 ECGs from 7397 patients. On both outcomes, acute coronary syndrome and occlusion myocardial infarction, the proposed decision fusion with retraining (Approach 2) achieved an AUROC of 0.878 (95% CI, 0.837–0.915) and 0.961 (95% CI, 0.934–0.983) on the test set and an average precision score of 0.749 (95% CI, 0.681–0.810) and 0.850 (95% CI, 0.763–0.918), respectively, outperforming all other proposed fusion and individual models. Conclusion: This study demonstrates the significance of fusion models that integrate complementary ECG representations to improve diagnostic accuracy in acute coronary events, highlighting a promising approach for real-time AI-augmented cardiovascular disease detection across diverse use cases.

Objective: Acute coronary syndrome (ACS) is a life-threatening condition requiring accurate diagnosis for better outcomes. However, variability in signs and symptoms among racial subgroups could cause disparities in diagnostic accuracy. In this study, we use machine learning models to diagnose ACS, focusing on mitigating disparities and ensuring fairness between Black and non-Black populations. Methods: We built on a state-of-the-art random forest classifier to compare three mitigation strategies. The first two approaches involved resampling or partitioning the data prior to training, while the third approach proposed an innovative framework called adversarial debiasing. To evaluate our model performance, we used the receiver operating characteristic (ROC) curve and an operating point at 80% specificity for clinical importance. Results: After mitigation with adversarial debiasing, the difference in sensitivities between the two subgroups decreased from 9.8% to 1.3%. Specifically, this approach achieved areas under the ROC of 0.810 and 0.817, and sensitivities of 70.1% and 71.4%, respectively for Black and non-Black subgroups. Conclusion: The proposed adversarial debiasing model outperformed the other two methods in both diagnostic accuracy and effectiveness in minimizing disparities. Significance: We expect this framework to achieve fair diagnostic models across diverse demographic populations globally and be generalizable to other outcomes.

Aims: Among patients with acute pulmonary embolism (PE), rapid identification of those with highest clinical risk can help guide life-saving treatment. However, current risk stratification algorithms involve a multistep process requiring physical exam, imaging, and laboratory results. We investigated the utility of electrocardiogram (ECG) alone to rapidly identify patients at elevated clinical risk by developing and validating a feature-based artificial intelligence (AI) model to predict clinical risk. Methods and results: Patients who were diagnosed with PE over a 9-year period, had an ECG within 1 day of presentation, and were evaluated by our PE response team (PERT) were included. A feature-based random forest model was trained to predict the PERT team’s risk stratification from the ECG alone. The ability of the model to predict the clinical risk categorization and the accuracy of both risk stratification approaches in predicting mortality were examined on a holdout test set. Of the overall cohort of 1376 patients, 55% had a submassive (intermediate risk) or massive (high risk) PE, which were grouped together as ‘severe PE’. The AI-ECG model was able to predict the clinical classification (low-risk vs. severe PE) with an AUC of 0.83 and F1 score of 0.78 in a holdout test set. A 30-day mortality and in-hospital mortality were significantly different between patients classified by the model as low vs. elevated risk. Conclusion: AI-based analysis of 12-lead ECGs may provide a useful tool in the risk stratification of PE, allowing for rapid identification and treatment of those at highest risk of adverse outcomes.

Objective: In this paper we develop and evaluate ECG-SMART-NET for occlusion myocardial infarction (OMI) identification. OMI is a severe form of heart attack characterized by complete blockage of one or more coronary arteries requiring immediate referral for cardiac catheterization to restore blood flow to the heart. Two thirds of OMI cases are difficult to visually identify from a 12-lead electrocardiogram (ECG) and can be potentially fatal if not identified quickly. Previous works on this topic are scarce, and current state-of-the-art evidence suggests both feature-based random forests and convolutional neural networks (CNNs) are promising approaches to improve ECG detection of OMI. Methods: While the ResNet architecture has been adapted for use with ECG recordings, it is not ideally suited to capture informative temporal features within each lead and the spatial concordance or discordance across leads. We propose a clinically informed modification of the ResNet-18 architecture. The model first learns temporal features through temporal convolutional layers with 1xk kernels followed by a spatial convolutional layer, after the residual blocks, with 12x1 kernels to learn spatial features. Results: ECG-SMART-NET was benchmarked against the original ResNet-18 and other state-of-the-art models on a multisite real-word clinical dataset that consists of 10, 393 ECGs from 7, 397 unique patients (rate of OMI = 7.2%). ECG-SMART-NET outperformed other models in the classification of OMI with a test AUC of 0.953 [0.921, 0.978]. Conclusion and Significance: ECG-SMART-NET can outperform the state-of-the-art random forest for OMI prediction and is better suited for this task than the original ResNet-18 architecture.

Immunotherapy has become a mainstay therapeutic in a variety of malignancies. Checkpoint inhibitors are used to enhance the endogenous immune response to target malignancy. The primary targets of checkpoint inhibitors include programed cell death receptor 1 (PD-1), programed cell death ligand 1 (PD-L1), and cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4). Despite clinical benefits, immune checkpoint inhibitors are associated with side effects termed immune-related adverse events (irAEs). We report a case of the PD-1 inhibitor (pembrolizumab) causing severe distal renal tubular acidosis (RTA). The patient presented with altered mental status and was found to have a severe non-anion gap metabolic acidosis after other potential etiologies were ruled out. We review the etiology of immune checkpoint inhibitor-induced RTA and review the case reports that have been described in the literature.

The incidence of cardiac arrhythmias is known to be associated with tissue heterogeneities including fibrosis. However, the impact of microscopic structural heterogeneities on conduction in excitable tissues remains poorly understood. In this study, we investigated how acellular microheterogeneities affect macroscopic conduction under conditions of normal and reduced excitability by utilizing a novel platform of paired in vitro and in silico studies to examine the mechanisms of conduction. Regular patterns of nonconductive micro-obstacles were created in confluent monolayers of the previously described engineered-excitable Ex293 cell line. Increasing the relative ratio of obstacle size to intra-obstacle strand width resulted in significant conduction slowing up to 23.6% and a significant increase in wavefront curvature anisotropy, a measure of spatial variation in wavefront shape. Changes in bulk electrical conductivity and in path tortuosity were insufficient to explain these observed macroscopic changes. Rather, microscale behaviors including local conduction slowing due to microscale branching, and conduction acceleration due to wavefront merging were shown to contribute to macroscopic phenomena. Conditions of reduced excitability led to further conduction slowing and a reversal of wavefront curvature anisotropy due to spatially non-uniform effects on microscopic slowing and acceleration. This unique experimental and computation platform provided critical mechanistic insights in the impact of microscopic heterogeneities on macroscopic conduction, pertinent to settings of fibrotic heart disease. It is well known that perturbations in the heart structure are associated with the initiation and maintenance of clinically significant cardiac arrhythmia. While previous studies have examined how single structural perturbations affect local electrical conduction, our understanding of how numerous microscopic heterogeneities act in aggregate to alter macroscopic electrical behavior is limited. In this study, we utilized simplified engineered excitable cells that contain the minimal machinery of excitability and can be directly computationally modeled. By pairing experimental and computational studies, we showed that the microscopic branching and collisions of electrical waves slow and speed conduction, respectively, resulting in macroscopic changes in the speed and pattern of electrical activation. These microscale behaviors are significantly altered under reduced excitability, resulting in exaggerated collision effects. Overall, this study helps improve our understanding of how microscopic structural heterogeneities in excitable tissue lead to abnormal action potential propagation, conducive to arrhythmias.

Cardiac arrhythmias have been traditionally simulated using continuous models that assume tissue homogeneity and use a relatively large spatial discretization. However, it is believed that the tissue fibrosis and collagen deposition, which occur on a micron-level, are critical factors in arrhythmogenesis in diseased tissues. Consequently, it remains unclear how well continuous models, which use averaged electrical properties, are able to accurately capture complex conduction behaviors such as re-entry in fibrotic tissues. The objective of this study was to compare re-entrant behavior in discrete microstructural models of fibrosis and in two types of equivalent continuous models, a homogenous continuous model and a hybrid continuous model with distinct heterogeneities. In the discrete model, increasing levels of tissue fibrosis lead to a substantial increase in the re-entrant cycle length which is inadequately reflected in the homogenous continuous models. These cycle length increases appear to be primarily due to increases in the tip path length and to altered restitution behavior, and suggest that it is critical to consider the discrete effects of fibrosis on conduction when studying arrhythmogenesis in fibrotic myocardium. Hybrid models are able to accurately capture some aspects of re-entry and, if carefully tuned, may provide a framework for simulating conduction in diseased tissues with both accuracy and efficiency.