Prospective Real-Time Validation of a Lung Ultrasound Deep Learning Model in the ICU.

OBJECTIVES: To evaluate the accuracy of a bedside, real-time deployment of a deep learning (DL) model capable of distinguishing between normal (A line pattern) and abnormal (B line pattern) lung parenchyma on lung ultrasound (LUS) in critically ill patients. DESIGN: Prospective, observational study evaluating the performance of a previously trained LUS DL model. Enrolled patients received a LUS examination with simultaneous DL model predictions using a portable device. Clip-level model predictions were analyzed and compared with blinded expert review for A versus B line pattern. Four prediction thresholding approaches were applied to maximize model sensitivity and specificity at bedside. SETTING: Academic ICU. PATIENTS: One-hundred critically ill patients admitted to ICU, receiving oxygen therapy, and eligible for respiratory imaging were included. Patients who were unstable or could not undergo an LUS examination were excluded. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: A total of 100 unique ICU patients (400 clips) were enrolled from two tertiary-care sites. Fifty-six patients were mechanically ventilated. When compared with gold standard expert annotation, the real-time inference yielded an accuracy of 95%, sensitivity of 93%, and specificity of 96% for identification of the B line pattern. Varying prediction thresholds showed that real-time modification of sensitivity and specificity according to clinical priorities is possible. CONCLUSIONS: A previously validated DL classification model performs equally well in real-time at the bedside when platformed on a portable device. As the first study to test the feasibility and performance of a DL classification model for LUS in a dedicated ICU environment, our results justify further inquiry into the impact of employing real-time automation of medical imaging into the care of the critically ill.

Fentanyl-Induced Block of hERG Channels Is Exacerbated by Hypoxia, Hypokalemia, Alkalosis, and the Presence of hERG1b.

Human ether-a-go-go-related gene (hERG) encodes the pore-forming subunit of the rapidly activating delayed rectifier potassium current (IKr) important for repolarization of cardiac action potentials. Drug-induced disruption of hERG channel function is a main cause of acquired long QT syndrome, which can lead to ventricular arrhythmias and sudden death. Illicit fentanyl use is associated with sudden death. We have demonstrated that fentanyl blocks hERG current (IhERG) at concentrations that overlap with the upper range of postmortem blood concentrations in fentanyl-related deaths. Since fentanyl can cause respiratory depression and electrolyte imbalances, in the present study we investigated whether certain pathologic circumstances exacerbate fentanyl-induced block of IhERG Our results show that chronic hypoxia or hypokalemia additively reduced IhERG with fentanyl. As well, high pH potentiated the fentanyl-mediated block of hERG channels, with an IC50 at pH 8.4 being 7-fold lower than that at pH 7.4. Furthermore, although the full-length hERG variant, hERG1a, has been widely used to study hERG channels, coexpression with the short variant, hERG1b (which does not produce current when expressed alone), produces functional hERG1a/1b channels, which gate more closely resembling native IKr Our results showed that fentanyl blocked hERG1a/1b channels with a 3-fold greater potency than hERG1a channels. Thus, in addition to a greater susceptibility due to the presence of hERG1b in the human heart, hERG channel block by fentanyl can be exacerbated by certain conditions, such as hypoxia, hypokalemia, or alkalosis, which may increase the risk of fentanyl-induced ventricular arrhythmias and sudden death. SIGNIFICANCE STATEMENT: This work demonstrates that heterologously expressed human ether a-go-go-related gene (hERG) 1a/1b channels, which more closely resemble rapidly activating delayed rectifier potassium current in the human heart, are blocked by fentanyl with a 3-fold greater potency than the previously studied hERG1a expressed alone. Additionally, chronic hypoxia, hypokalemia, and alkalosis can increase the block of hERG current by fentanyl, potentially increasing the risk of cardiac arrhythmias and sudden death.

Blockade of the Human Ether A-Go-Go-Related Gene (hERG) Potassium Channel by Fentanyl.

The human ether-a-go-go-related gene (hERG) encodes the pore-forming subunit of the rapidly activating delayed rectifier potassium channel (IKr). Drug-mediated or medical condition-mediated disruption of hERG function is the primary cause of acquired long-QT syndrome, which predisposes affected individuals to ventricular arrhythmias and sudden death. Fentanyl abuse poses a serious health concern, with abuse and death rates rising over recent years. As fentanyl has a propensity to cause sudden death, we investigated its effects on the hERG channel. The effects of norfentanyl, the main metabolite, and naloxone, an antidote used in fentanyl overdose, were also examined. Currents of hERG channels stably expressed in HEK293 cells were recorded using the whole-cell voltage-clamp method. When hERG tail currents were analyzed upon -50 mV repolarization after a 50 mV depolarization, fentanyl and naloxone blocked hERG current (IhERG) with IC50 values of 0.9 and 74.3 µM, respectively, whereas norfentanyl did not block. However, fentanyl-mediated block of IhERG was voltage dependent. When a voltage protocol that mimics a human ventricular action potential (AP) was used, fentanyl blocked IhERG with an IC50 of 0.3 µM. Furthermore, fentanyl (0.5 µM) prolonged AP duration and blocked IKr in ventricular myocytes isolated from neonatal rats. The concentrations of fentanyl used in this study were higher than seen with clinical use but overlap with postmortem overdose concentrations. Although mechanisms of fentanyl-related sudden death need further investigation, blockade of hERG channels may contribute to the death of individuals with high-concentration overdose or compromised cardiac repolarization.

Differential Regulation of Human Ether-à-Go-Go-Related Gene (hERG) Current and Expression by Activation of Protein Kinase C.

The human ether-à-go-go-related gene (hERG) encodes the channel that conducts the rapidly activating delayed rectifier potassium current (IKr) in the heart. Reduction in IKr causes long QT syndrome, which can lead to fatal arrhythmias triggered by stress. One potential link between stress and hERG function is protein kinase C (PKC) activation; however, seemingly conflicting results regarding PKC regulation of hERG have been reported. We investigated the effects of PKC activation using phorbol 12-myristate 13-acetate (PMA) on hERG channels expressed in human embryonic kidney cell line 293 (HEK293) cells and IKr in isolated neonatal rat ventricular myocytes. Acute activation of PKC by PMA (30 nM, 30 minutes) reduced both hERG current (IhERG) and IKr Chronic activation of PKC by PMA (30 nM, 16 hours) increased IKr in cardiomyocytes and the expression level of hERG proteins; however, chronic (30 nM, 16 hours) PMA treatment decreased IhERG, which became larger than untreated control IhERG after PMA removal for 4 hours. Deletion of amino acid residues 2-354 (Δ2-354 hERG) or 1-136 of the N terminus (ΔN 136 hERG) abolished acute PMA (30 nM, 30 minutes)-mediated IhERG reduction. In contrast to wild-type hERG channels, chronic activation of PKC by PMA (30 nM, 16 hours) increased both Δ2-354 hERG and ΔN136 hERG expression levels and currents. The increase in hERG protein was associated with PKC-induced phosphorylation (inhibition) of Nedd4-2, an E3 ubiquitin ligase that mediates hERG degradation. We conclude that PKC regulates hERG in a balanced manner, increasing expression through inhibiting Nedd4-2 while decreasing current through targeting a site(s) within the N terminus.

The N terminus and transmembrane segment S1 of Kv1.5 can coassemble with the rest of the channel independently of the S1-S2 linkage.

The voltage-gated potassium channel Kv1.5 belongs to the Shaker superfamily. Kv1.5 is composed of four subunits, each comprising 613 amino acids, which make up the N terminus, six transmembrane segments (S1-S6), and the C terminus. We recently demonstrated that, in HEK cells, extracellularly applied proteinase K (PK) cleaves Kv1.5 channels at a single site in the S1-S2 linker. This cleavage separates Kv1.5 into an N-fragment (N terminus to S1) and a C-fragment (S2 to C terminus). Interestingly, the cleavage does not impair channel function. Here, we investigated the role of the N terminus and S1 in Kv1.5 expression and function by creating plasmids encoding various fragments, including those that mimic PK-cleaved products. Our results disclosed that although expression of the pore-containing fragment (Frag(304-613)) alone could not produce current, coexpression with Frag(1-303) generated a functional channel. Immunofluorescence and biotinylation analyses uncovered that Frag(1-303) was required for Frag(304-613) to traffic to the plasma membrane. Biochemical analysis revealed that the two fragments interacted throughout channel trafficking and maturation. In Frag(1-303)+(304-613)-coassembled channels, which lack a covalent linkage between S1 and S2, amino acid residues 1-209 were important for association with Frag(304-613), and residues 210-303 were necessary for mediating trafficking of coassembled channels to the plasma membrane. We conclude that the N terminus and S1 of Kv1.5 can attract and coassemble with the rest of the channel (i.e. Frag(304-613)) to form a functional channel independently of the S1-S2 linkage.

Automated Real-Time Detection of Lung Sliding Using Artificial Intelligence: A Prospective Diagnostic Accuracy Study.

BACKGROUND: Rapid evaluation for pneumothorax is a common clinical priority. Although lung ultrasound (LUS) often is used to assess for pneumothorax, its diagnostic accuracy varies based on patient and provider factors. To enhance the performance of LUS for pulmonary pathologic features, artificial intelligence (AI)-assisted imaging has been adopted; however, the diagnostic accuracy of AI-assisted LUS (AI-LUS) deployed in real time to diagnose pneumothorax remains unknown. RESEARCH QUESTION: In patients with suspected pneumothorax, what is the real-time diagnostic accuracy of AI-LUS to recognize the absence of lung sliding? STUDY DESIGN AND METHODS: We performed a prospective AI-assisted diagnostic accuracy study of AI-LUS to recognize the absence of lung sliding in a convenience sample of patients with suspected pneumothorax. After calibrating the model parameters and imaging settings for bedside deployment, we prospectively evaluated its diagnostic accuracy for lung sliding compared with a reference standard of expert consensus. RESULTS: Two hundred forty-one lung sliding evaluations were derived from 62 patients. AI-LUS showed a sensitivity of 0.921 (95% CI, 0.792-0.973), specificity of 0.802 (95% CI, 0.735-0.856), area under the receiver operating characteristic curve of 0.885 (95% CI, 0.828-0.956), and accuracy of 0.824 (95% CI, 0.766-0.870) for the diagnosis of absent lung sliding. INTERPRETATION: In this study, real-time AI-LUS showed high sensitivity and moderate specificity to identify the absence of lung sliding. Further research to improve model performance and optimize the integration of AI-LUS into existing diagnostic pathways is warranted.

Improving the Generalizability and Performance of an Ultrasound Deep Learning Model Using Limited Multicenter Data for Lung Sliding Artifact Identification.

Deep learning (DL) models for medical image classification frequently struggle to generalize to data from outside institutions. Additional clinical data are also rarely collected to comprehensively assess and understand model performance amongst subgroups. Following the development of a single-center model to identify the lung sliding artifact on lung ultrasound (LUS), we pursued a validation strategy using external LUS data. As annotated LUS data are relatively scarce-compared to other medical imaging data-we adopted a novel technique to optimize the use of limited external data to improve model generalizability. Externally acquired LUS data from three tertiary care centers, totaling 641 clips from 238 patients, were used to assess the baseline generalizability of our lung sliding model. We then employed our novel Threshold-Aware Accumulative Fine-Tuning (TAAFT) method to fine-tune the baseline model and determine the minimum amount of data required to achieve predefined performance goals. A subgroup analysis was also performed and Grad-CAM++ explanations were examined. The final model was fine-tuned on one-third of the external dataset to achieve 0.917 sensitivity, 0.817 specificity, and 0.920 area under the receiver operator characteristic curve (AUC) on the external validation dataset, exceeding our predefined performance goals. Subgroup analyses identified LUS characteristics that most greatly challenged the model's performance. Grad-CAM++ saliency maps highlighted clinically relevant regions on M-mode images. We report a multicenter study that exploits limited available external data to improve the generalizability and performance of our lung sliding model while identifying poorly performing subgroups to inform future iterative improvements. This approach may contribute to efficiencies for DL researchers working with smaller quantities of external validation data.

Introducing medical students to deep learning through image labelling: a new approach to meet calls for greater artificial intelligence fluency among medical trainees.

Our approach addresses the urgent need for AI experience for the doctors of tomorrow. Through a medical education-focused approach to data labelling, we have fostered medical student competence in medical imaging and AI. We envision our framework being applied at other institutions and academic groups to develop robust labelling programs for research endeavours. Application of our approach to core visual modalities within medicine (e.g. interpretation of ECGs, diagnostic imaging, dermatologic findings) can lead to valuable student experience and competence in domains that feature prominently in clinical practice, while generating much needed data in fields that are ripe for AI integration.

Notre approche répond au besoin urgent de familiariser les médecins de demain avec l'IA. Nous avons cherché à développer leurs compétences en imagerie médicale et en IA par une approche à l'étiquetage de données axée sur la formation médicale. D'autres établissements et groupes universitaires souhaitant mettre sur pied des programmes d'étiquetage solides pour leurs projets de recherche pourraient adopter notre modèle. L'application de notre approche aux principales modalités visuelles en médecine (par exemple, l'interprétation des ECG, l'imagerie diagnostique, le diagnostic des lésions dermatologiques) peut permettre aux étudiants d'acquérir une expérience et des compétences précieuses dans des domaines importants de la pratique clinique, tout en procurant des données indispensables dans des secteurs qui sont mûrs pour une intégration de l'IA.

Accurate assessment of the lung sliding artefact on lung ultrasonography using a deep learning approach.

Pneumothorax is a potentially life-threatening condition that can be rapidly and accurately assessed via the lung sliding artefact generated using lung ultrasound (LUS). Access to LUS is challenged by user dependence and shortage of training. Image classification using deep learning methods can automate interpretation in LUS and has not been thoroughly studied for lung sliding. Using a labelled LUS dataset from 2 academic hospitals, clinical B-mode (also known as brightness or two-dimensional mode) videos featuring both presence and absence of lung sliding were transformed into motion (M) mode images. These images were subsequently used to train a deep neural network binary classifier that was evaluated using a holdout set comprising 15% of the total data. Grad-CAM explanations were examined. Our binary classifier using the EfficientNetB0 architecture was trained using 2535 LUS clips from 614 patients. When evaluated on a test set of data uninvolved in training (540 clips from 124 patients), the model performed with a sensitivity of 93.5%, specificity of 87.3% and an area under the receiver operating characteristic curve (AUC) of 0.973. Grad-CAM explanations confirmed the model's focus on relevant regions on M-mode images. Our solution accurately distinguishes between the presence and absence of lung sliding artefacts on LUS.

Enhancing Annotation Efficiency with Machine Learning: Automated Partitioning of a Lung Ultrasound Dataset by View.

BACKGROUND: Annotating large medical imaging datasets is an arduous and expensive task, especially when the datasets in question are not organized according to deep learning goals. Here, we propose a method that exploits the hierarchical organization of annotating tasks to optimize efficiency. METHODS: We trained a machine learning model to accurately distinguish between one of two classes of lung ultrasound (LUS) views using 2908 clips from a larger dataset. Partitioning the remaining dataset by view would reduce downstream labelling efforts by enabling annotators to focus on annotating pathological features specific to each view. RESULTS: In a sample view-specific annotation task, we found that automatically partitioning a 780-clip dataset by view saved 42 min of manual annotation time and resulted in 55±6 additional relevant labels per hour. CONCLUSIONS: Automatic partitioning of a LUS dataset by view significantly increases annotator efficiency, resulting in higher throughput relevant to the annotating task at hand. The strategy described in this work can be applied to other hierarchical annotation schemes.

Automation of Lung Ultrasound Interpretation via Deep Learning for the Classification of Normal versus Abnormal Lung Parenchyma: A Multicenter Study.

Lung ultrasound (LUS) is an accurate thoracic imaging technique distinguished by its handheld size, low-cost, and lack of radiation. User dependence and poor access to training have limited the impact and dissemination of LUS outside of acute care hospital environments. Automated interpretation of LUS using deep learning can overcome these barriers by increasing accuracy while allowing point-of-care use by non-experts. In this multicenter study, we seek to automate the clinically vital distinction between A line (normal parenchyma) and B line (abnormal parenchyma) on LUS by training a customized neural network using 272,891 labelled LUS images. After external validation on 23,393 frames, pragmatic clinical application at the clip level was performed on 1162 videos. The trained classifier demonstrated an area under the receiver operating curve (AUC) of 0.96 (±0.02) through 10-fold cross-validation on local frames and an AUC of 0.93 on the external validation dataset. Clip-level inference yielded sensitivities and specificities of 90% and 92% (local) and 83% and 82% (external), respectively, for detecting the B line pattern. This study demonstrates accurate deep-learning-enabled LUS interpretation between normal and abnormal lung parenchyma on ultrasound frames while rendering diagnostically important sensitivity and specificity at the video clip level.