Ernest V Garcia1, J Larry Klein2, Valeria Moncayo3, C David Cooke3,4, Christian Del'Aune4, Russell Folks3, Liudmila Verdes Moreiras3, Fabio Esteves3. 1. Department of Radiology and Imaging Sciences, Emory University, 101 Woodruff Circle, Room 1203, Atlanta, GA, 30322, USA. ernest.garcia@emory.edu. 2. Division of Cardiology, Department of Medicine, UNC School of Medicine, University of North Carolina, Chapel Hill, NC, USA. 3. Department of Radiology and Imaging Sciences, Emory University, 101 Woodruff Circle, Room 1203, Atlanta, GA, 30322, USA. 4. Syntermed, Inc., Atlanta, GA, USA.
Abstract
OBJECTIVES: To describe and validate an artificial intelligence (AI)-driven structured reporting system by direct comparison of automatically generated reports to results from actual clinical reports generated by nuclear cardiology experts. BACKGROUND: Quantitative parameters extracted from myocardial perfusion imaging (MPI) studies are used by our AI reporting system to generate automatically a guideline-compliant structured report (sR). METHOD: A new nonparametric approach generates distribution functions of rest and stress, perfusion, and thickening, for each of 17 left ventricle segments that are then transformed to certainty factors (CFs) that a segment is hypoperfused, ischemic. These CFs are then input to our set of heuristic rules used to reach diagnostic findings and impressions propagated into a sR referred as an AI-driven structured report (AIsR). The diagnostic accuracy of the AIsR for detecting coronary artery disease (CAD) and ischemia was tested in 1,000 patients who had undergone rest/stress SPECT MPI. RESULTS: At the high-specificity (SP) level, in a subset of 100 patients, there were no statistical differences in the agreements between the AIsr, and nine experts' impressions of CAD (P = .33) or ischemia (P = .37). This high-SP level also yielded the highest accuracy across global and regional results in the 1,000 patients. These accuracies were statistically significantly better than the other two levels [sensitivity (SN)/SP tradeoff, high SN] across all comparisons. CONCLUSIONS: This AI reporting system automatically generates a structured natural language report with a diagnostic performance comparable to those of experts.
OBJECTIVES: To describe and validate an artificial intelligence (AI)-driven structured reporting system by direct comparison of automatically generated reports to results from actual clinical reports generated by nuclear cardiology experts. BACKGROUND: Quantitative parameters extracted from myocardial perfusion imaging (MPI) studies are used by our AI reporting system to generate automatically a guideline-compliant structured report (sR). METHOD: A new nonparametric approach generates distribution functions of rest and stress, perfusion, and thickening, for each of 17 left ventricle segments that are then transformed to certainty factors (CFs) that a segment is hypoperfused, ischemic. These CFs are then input to our set of heuristic rules used to reach diagnostic findings and impressions propagated into a sR referred as an AI-driven structured report (AIsR). The diagnostic accuracy of the AIsR for detecting coronary artery disease (CAD) and ischemia was tested in 1,000 patients who had undergone rest/stress SPECT MPI. RESULTS: At the high-specificity (SP) level, in a subset of 100 patients, there were no statistical differences in the agreements between the AIsr, and nine experts' impressions of CAD (P = .33) or ischemia (P = .37). This high-SP level also yielded the highest accuracy across global and regional results in the 1,000 patients. These accuracies were statistically significantly better than the other two levels [sensitivity (SN)/SP tradeoff, high SN] across all comparisons. CONCLUSIONS: This AI reporting system automatically generates a structured natural language report with a diagnostic performance comparable to those of experts.
Authors: Ernest V Garcia; Andrew Taylor; Russell Folks; Daya Manatunga; Raghuveer Halkar; Bital Savir-Baruch; Eva Dubovsky Journal: Eur J Nucl Med Mol Imaging Date: 2012-05-30 Impact factor: 9.236
Authors: Christopher L Hansen; Richard A Goldstein; Olakunle O Akinboboye; Daniel S Berman; Elias H Botvinick; Keith B Churchwell; C David Cooke; James R Corbett; S James Cullom; Seth T Dahlberg; Regina S Druz; Edward P Ficaro; James R Galt; Ravi K Garg; Guido Germano; Gary V Heller; Milena J Henzlova; Mark C Hyun; Lynne L Johnson; April Mann; Benjamin D McCallister; Robert A Quaife; Terrence D Ruddy; Senthil N Sundaram; Raymond Taillefer; R Parker Ward; John J Mahmarian Journal: J Nucl Cardiol Date: 2007 Nov-Dec Impact factor: 5.952
Authors: Andrew Taylor; Andrew N Hill; José N E Binongo; Amita K Manatunga; Raghuveer Halkar; Eva V Dubovsky; Ernest V Garcia Journal: AJR Am J Roentgenol Date: 2007-05 Impact factor: 3.959
Authors: Pamela S Douglas; Robert C Hendel; Jennifer E Cummings; John M Dent; John McB Hodgson; Udo Hoffmann; Robert J Horn; W Gregory Hundley; Charles E Kahn; Gerard R Martin; Frederick A Masoudi; Eric D Peterson; Geoffrey L Rosenthal; Harry Solomon; Arthur E Stillman; Shawn D Teague; James D Thomas; Peter L Tilkemeier; Wm Guy Weigold Journal: J Am Coll Cardiol Date: 2009-01-06 Impact factor: 24.094
Authors: Fabio P Esteves; James R Galt; Russell D Folks; Liudmila Verdes; Ernest V Garcia Journal: J Nucl Cardiol Date: 2013-11-28 Impact factor: 5.952
Authors: Fabio P Esteves; Paolo Raggi; Russell D Folks; Zohar Keidar; J Wells Askew; Shmuel Rispler; Michael K O'Connor; Liudmilla Verdes; Ernest V Garcia Journal: J Nucl Cardiol Date: 2009-08-18 Impact factor: 5.952