No abstract

While data-driven approaches excel at many image analysis tasks, the performance of these approaches is often limited by a shortage of annotated data available for training. Recent work in semi-supervised learning has shown that meaningful representations of images can be obtained from training with large quantities of unlabeled data, and that these representations can improve the performance of supervised tasks. Here, we demonstrate that an unsupervised jigsaw learning task, in combination with supervised training, results in up to a 9.8% improvement in correctly classifying lesions in colonoscopy images when compared to a fully-supervised baseline. We additionally benchmark improvements in domain adaptation and out-of-distribution detection, and demonstrate that semi-supervised learning outperforms supervised learning in both cases. In colonoscopy applications, these metrics are important given the skill required for endoscopic assessment of lesions, the wide variety of endoscopy systems in use, and the homogeneity that is typical of labeled datasets.

Aims Post-infarct myocardium contains viable corridors traversing scar or lipomatous metaplasia (LM). Ventricular tachycardia (VT) circuitry has been separately reported to associate with corridors that traverse LM and with repolarization heterogeneity. We examined the association of corridor activation recovery interval (ARI) and ARI dispersion with surrounding tissue type. Methods and results The cohort included 33 post-infarct patients from the prospective Intra-Myocardial Fat Deposition and Ventricular Tachycardia in Cardiomyopathy (INFINITY) study. We co-registered scar and corridors from late gadolinium enhanced magnetic resonance, and LM from computed tomography with intracardiac electrogram locations. Activation recovery interval was calculated during sinus or ventricular pacing, as the time interval from the minimum derivative within the QRS to the maximum derivative within the T-wave on unipolar electrograms. Regional ARI dispersion was defined as the standard deviation (SD) of ARI per AHA segment (ARISD). Lipomatous metaplasia exhibited higher ARI than scar [325 (interquartile range 270–392) vs. 313 (255–374), P < 0.001]. Corridors critical to VT re-entry were more likely to traverse through or near LM and displayed prolonged ARI compared with non-critical corridors [355 (319–397) vs. 302 (279–333) ms, P < 0.001]. ARISD was more closely associated with LM than with scar (likelihood ratio χ2 50 vs. 12, and 4.2-unit vs. 0.9-unit increase in 0.01*Log(ARISD) per 1 cm2 increase per AHA segment). Additionally, LM and scar exhibited interaction (P < 0.001) in their association with ARISD. Conclusion Lipomatous metaplasia is closely associated with prolonged local action potential duration of corridors and ARI dispersion, which may facilitate the propensity of VT circuit re-entry.

Catheter ablation is increasingly utilized to treat patients with atrial fibrillation (AF). Despite progress in technology and procedural strategy, there remain significant limitations with suboptimal outcomes. The role of imaging has continued to evolve, and multimodality imaging now presents an important opportunity to make substantial progress in the safety and efficacy of ablation. In this review, we discuss the history of imaging in the ablation of AF with a specific focus on the ability of cardiac computed tomography and magnetic resonance imaging to characterize anatomy, arrhythmogenic substrate, and guide ablation strategy. We will review the progress that has been made and highlight many of the limitations as well as future directions for the field.

Introduction Esophageal injury during atrial fibrillation (AF) ablation is a life‐threatening complication. We sought to measure the association of esophageal temperature attenuation with radiofrequency (RF) electrode impedance, contact force, and distance from the esophagus. Methods The retrospective study cohort included 35 patients with mean age 64 ± 10 years, of whom 74.3% were male, and 40% had persistent AF. All patients had undergone preprocedural cardiac magnetic resonance (CMR) followed by AF ablation with luminal esophageal temperature monitoring. Lesion locations were co‐registered with CMR image segmentations of left atrial and esophageal anatomy. Luminal esophageal temperature, time matched RF lesion data, and ablation distance from the nearest esophageal location were collected as panel data. Results Luminal esophageal temperature changes corresponding to 3667 distinct lesions, delivered with mean power 27.9 ± 5.5 W over a mean duration of 22.2 ± 10.5 s were analyzed. In multivariable analyses, clustered per patient, examining posterior wall lesions only, and adjusted for lesion power and duration as set by the operator, lesion distance from the esophagus (−0.003°C/mm, p < .001), and baseline impedance (−0.015°C/Ω, p < .001) were associated with changes in luminal esophageal temperature. Conclusion Esophageal luminal temperature rises are associated with shorter lesion distance from esophagus and lower baseline impedance during RF lesion delivery. When procedural strategy requires RF delivery near the esophagus, selection of sites with higher baseline impedance may improve safety.