Diagnostic accuracy of an artificial neural network compared with statistical quantitation of myocardial perfusion images: a Japanese multicenter study

Abstract

PurposeArtificial neural networks (ANN) might help to diagnose coronary artery disease. This study aimed to determine whether the diagnostic accuracy of an ANN-based diagnostic system and conventional quantitation are comparable.MethodsThe ANN was trained to classify potentially abnormal areas as true or false based on the nuclear cardiology expert interpretation of 1001 gated stress/rest 99mTc-MIBI images at 12 hospitals. The diagnostic accuracy of the ANN was compared with 364 expert interpretations that served as the gold standard of abnormality for the validation study. Conventional summed stress/rest/difference scores (SSS/SRS/SDS) were calculated and compared with receiver operating characteristics (ROC) analysis.ResultsThe ANN generated a better area under the ROC curves (AUC) than SSS (0.92 vs. 0.82, p < 0.0001), indicating better identification of stress defects. The ANN also generated a better AUC than SDS (0.90 vs. 0.75, p < 0.0001) for stress-induced ischemia. The AUC for patients with old myocardial infarction based on rest defects was 0.97 (0.91 for SRS, p = 0.0061), and that for patients with and without a history of revascularization based on stress defects was 0.94 and 0.90 (p = 0.0055 and p < 0.0001 vs. SSS, respectively). The SSS/SRS/SDS steeply increased when ANN values (probability of abnormality) were >0.80.ConclusionThe ANN was diagnostically accurate in various clinical settings, including that of patients with previous myocardial infarction and coronary revascularization. The ANN could help to diagnose coronary artery disease.

Artificial neural network retrained to detect myocardial ischemia using a Japanese multicenter database

“…The practical method of applying the ANN to clinical practice should be considered. The relationship between ANN probability and defect scores is not linear [ 3 , 4 ]. Summed stress, rest and difference scores all steeply increased when the ANN probability was > 0.80, which means that the ANN probability could play a unique role in the diagnosis of coronary artery disease.…”

“…The first version of the ANN was trained on data from 1,051 Swedish patients (male: 47%; age, 62 ± 10 years) and experienced Swedish physicians classified perfusion as normal or defective [ 2 ]. Twelve hospitals in Japan collaborated to train version 1.1 ( n = 1,001 patients; 75% male; 69 ± 10 years) using 99m Tc-MIBI as the tracer [ 4 ]. At least two Japanese nuclear cardiology experts determined abnormal stress defects and stress-induced ischemia by consensus.…”

“…That ANN was trained to detect myocardial stress perfusion defects and induced ischemia on a Swedish database, but its diagnostic ability was comparable to that of expert interpretation for Japanese patients. Thereafter, the diagnostic ability was further improved by training the ANN on a Japanese multicenter database ( n = 1,001) using 99m Tc-methoxyisobutylisonitrile (MIBI) MPI [ 4 ]. That validation study indicated that the ANN had good diagnostic ability comparable to nuclear cardiology expert interpretation, as the area under the receiver-operating characteristics (ROC) curve (AUC) was 0.92.…”

Usefulness of an artificial neural network for a beginner to achieve similar interpretations to an expert when examining myocardial perfusion images

“…1. [6,7] Usually, a 17-segment model is used for myocardial perfusion scoring. [8] However, our study aimed to examine the effects of AI on the performance of beginners during image interpretation.…”

Machine learning for nuclear cardiology: The way forward

“…Several recent studies in Nuclear Cardiology [7][8][9][10][11][12][13] have attempted to use machine learning in the identification of perfusion defects and location, prediction of revascularization and cardiovascular events, and diagnostic and prognostic accuracy (Table 2). However, viewing through the lens of these studies, the investigation by Alonso et al, is singular to apply supervised machine learning in predicting cardiac death from the features while attempting to identify interpretable highperformance model.…”