Our previous work with 123iodine metaiodobenzylguanidine (123I-mIBG) radionuclide imaging among patients with cardiomyopathy reported limitations associated with theprognostic power of global parameters derived from planar imaging. Employing multivariate analysis, we further showed the regionalwashout associated with territories adjacent to infarcted myocardiumobtained from single-photon emission computed tomography imaging (SPECT) yielded superior prognostic power over the other planarand SPECT indices in predicting future cardiac events . The aimof this study was to apply an artificial neural network (Neural Analyser version 2.9.5) to the original data from the same patient cohortto evaluate the most potent prognostic index for future cardiac eventsamong patient with cardiomyopathy.Methods: The original data were reevaluated using an artificial neural network (Neural Analyser version 2.9.5). There were 84 inputvariables in the original 22 patients from clinical data, electrocardiogram (rest, stress, and continuous ambulatory electrocardiogramrecording), transthoracic echocardiography, coronary angiogram, sestamibi myocardial perfusion SPECT, planar and SPECT 123ImIBG, and genetic and biomarkers, detailed in the previous work.A single binary output was a cardiac event or no cardiac event inthe follow-up period.Results: Following training and validation phases, the optimalnumber of inputs was determined to be two with a training lossof 0.025 and selection loss <0.001. The final architecture had inputs of a change in left ventricular ejection fraction (D > 10%)and 123I-mIBG planar global washout (>30%), two hidden layersof 6 and 1 node, respectively, and a binary output. Using receiveroperator characteristics analysis demonstrated an area under thecurve of 0.75 correlating to a sensitivity of 100% and specificityof 50%.Conclusion: The premise that regional washout of 123I-mIBGSPECT from noninfarcted tissue is the best predictor of cardiacevents was built on has a sound and logical foundation. By artificialneural network analysis; however, 123I-mIBG planar global washoutof >30% was shown to be the best indicator for risk of cardiacevent when accompanied by a decline in left ventricular ejection fraction of >10%. Further investigation should be undertaken assessingassimilation into big data and the potential for automated featureextraction from raw image datasets with convolutional neuralnetworks.
Currie, G., Iqbal, B., & Kiat, H. (2019). Intelligent imaging: Radiomics and artificial neural networks in heart failure. Journal of Medical Imaging and Radiation Sciences, 50(4), 571-574. https://doi.org/10.1016/j.jmir.2019.08.006