Spatio-temporal hybrid neural networks reduce erroneous human “judgement calls” in the diagnosis of Takotsubo syndrome

Abstract

Background: We investigate whether deep learning (DL) neural networks can reduce erroneous human "judgment calls" on bedside echocardiograms and help distinguish Takotsubo syndrome (TTS) from anterior wall ST segment elevation myocardial infarction (STEMI). Methods: We developed a single-channel (DCNN[2D SCI]), a multi-channel (DCNN[2D MCI]), and a 3-dimensional (DCNN[2D+t]) deep convolution neural network, and a recurrent neural network (RNN) based on 17,280 still-frame images and 540 videos from 2-dimensional echocardiograms in 10 years (1 January 2008 to 1 January 2018) retrospective cohort in University of Iowa (UI) and eight other medical centers. Echocardiograms from 450 UI patients were randomly divided into training and testing sets for internal training, testing, and model construction. Echocardiograms of 90 patients from the other medical centers were used for external validation to evaluate the model generalizability. A total of 49 board-certified human readers performed human-side classification on the same echocardiography dataset to compare the diagnostic performance and help data visualization. Findings: The DCNN (2D SCI), DCNN (2D MCI), DCNN(2D+t), and RNN models established based on UI dataset for TTS versus STEMI prediction showed mean diagnostic accuracy 73%, 75%, 80%, and 75% respectively, and mean diagnostic accuracy of 74%, 74%, 77%, and 73%, respectively, on the external validation. DCNN(2D+t) (area under the curve [AUC] 0¢787 vs. 0¢699, P = 0¢015) and RNN models (AUC 0¢774 vs. 0¢699, P = 0¢033) outperformed human readers in differentiating TTS and STEMI by reducing human erroneous judgement calls on TTS. Interpretation: Spatio-temporal hybrid DL neural networks reduce erroneous human "judgement calls" in distinguishing TTS from anterior wall STEMI based on bedside echocardiographic videos.

Using artificial intelligence in the development of diagnostic models of coronary artery disease with imaging markers: A scoping review

“…Approximately, half of the included studies were conducted in years 2020 (12/46, 26.0%) ( 14 , 15 , 17 , 19 , 20 , 25 , 26 , 30 , 36 , 37 , 45 , 56 ) and 2021 (12/46, 26.0%) ( 13 , 16 , 21 , 24 , 29 , 33 , 39 , 40 , 43 , 46 , 50 , 53 ) ( Figure 2 ). The corresponding authors of the included studies were from 13 countries, including the United States (14/46, 30.3%) ( 12 , 16 , 21 , 22 , 29 , 31 , 34 , 36 , 39 , 41 , 42 , 48 , 49 , 53 ), China (11/46, 23.9%) ( 15 , 19 , 20 , 35 , 37 , 38 , 40 , 45 , 46 , 52 , 54 ), Japan (8/46, 17.3%) ( 17 , 23 , 25 , 27 , 28 , 32 , 55 , 56 ), Greece (2/46, 4.3%) ( 13 , 14 ), and Netherlands (2/46, 4.3%) ( 44 , 47 ), whereas Italy ( 26 ), Canada ( 11 ), India ( 30 ), Ko...…”

“…For data sources, private data (data collected by centers) (35/46, 76.0%) ( 11 – 21 , 23 , 24 , 27 , 33 – 50 , 53 , 54 , 56 ) were the most commonly used data sources for the development of AI models. Except for one article for which the data source is unclear ( 28 ), public data (10/46, 21.7%) were the other sources of data for AI models ( 22 , 25 , 26 , 29 – 32 , 51 , 52 , 55 ).…”

“…Except for one article for which the data source is unclear ( 28 ), public data (10/46, 21.7%) were the other sources of data for AI models ( 22 , 25 , 26 , 29 – 32 , 51 , 52 , 55 ). Most studies were single-center studies, accounting for 76.0% (35/46), and 19.5% (9/46) were multi-center studies ( 16 , 20 , 29 , 33 , 41 , 42 , 51 , 53 , 55 ). There were three major types of study designs: cohort study (30/46, 65.2%) ( 11 – 28 , 36 – 38 , 40 – 43 , 46 – 50 ), case–control (8/46, 17.3%) ( 34 , 35 , 39 , 44 , 52 – 54 , 56 ), and nested case–control (1/46, 2.1%) ( 33 ), whereas for the other 15.3% of the studies (7/46) ( 29 – 32 , 45 , 51 , 55 ), the type of study design could not be determined based on the information in the article.…”

“…Reference standards for determining the outcomes were only mentioned in 36 of the 46 studies. Experts (11/46, 23.9%) ( 11 , 16 , 20 , 21 , 36 , 38 , 39 , 45 , 48 , 49 , 52 ), such as cardiologists or radiologists, and coronary angiography (13/46, 28.3%) ( 12 – 15 , 17 – 19 , 33 , 41 – 43 , 46 , 53 ) were the two main reference standards. Coronary angiograms and experienced physicians (6/46, 13.0%) ( 28 , 29 , 31 , 32 , 51 , 55 ), fractional flow reserve (FFR) (4/46, 8.6%) ( 34 , 35 , 44 , 47 ), and clinical characteristics, electrocardiogram, and laboratory test index (2/46, 4.3%) ( 40 , 54 ) were used as the reference standards for CAD in other studies.…”

“…The most common medical imaging used was computed tomography (CT), comprising 73.9% (34/46) of the studies, which included single-photon emission computed tomography (SPECT) (15/46, 32.6%) ( 12 – 14 , 17 , 18 , 21 , 27 – 29 , 31 , 32 , 41 , 42 , 51 , 55 ), coronary computed tomography angiography (CCTA) (15/46, 32.6%) ( 15 , 16 , 19 , 23 , 26 , 30 , 34 – 36 , 39 , 40 , 44 , 46 – 48 ), optical coherence tomography (OCT) (3/46, 6.5%) ( 11 , 37 , 38 ), and non-contrast CT (1/46, 2.1%) ( 49 ). Other more commonly used medical imaging techniques were ultrasonography (5/46, 10.8%) ( 22 , 24 , 33 , 53 , 56 ), MR (2/46,4.3%) ( 52 , 54 ), and X-ray (2/46, 4.3%) ( 43 , 45 ). In contrast, the least commonly used images were coronary angioscopy (1/46, 2.1%) ( 25 ), histological slides (1/46, 2.1%) ( 50 ), and facial photo (1/46, 2.1%) ( 20 ).…”