Comparing Artificial Intelligence AI-Rad Companion Chest CT vs Experts on Emphysema, Lung Nodules, Thoracic Aorta and Thoracic Spine

Résumé

Background

The objective was to evaluate the performance of AI-Rad Companion software (AIRC-cCT)(Siemens Healthineers®) by analyzing chest CT scans and comparing the results against Radiologists.

Methods

AIRC-cCT retrospectively assessed 348 chest CTs, evaluating emphysema, lung nodules, thoracic aorta diameters and thoracic vertebral spine. An Expert Panel reviewed all results/reportings and was considered the ground truth.

Results

AIRC-cCT was adequately sensitive(68,5%) and specific(64,7%) detecting emphysema. The software misinterpreted low density areas as emphysema but offers the advantage of quantification emphysema.

 The 6 biggest nodules presented by AIRC-cCT were considered, with a total of 1003. Expert Panel validated 677(67,5%) as real nodules. 209 nodules(20,8%) were regarded as non-relevant(125 granulomas, 84 perifissural nodules, among others).

AIRC-cCT detected 101 scans with pathological dilation of the Aorta. Expert Panel and AIRC-cCT strongly agreed about aortic dilation status(Cohen’s kappa = 98,9%, 95% CI 98,1–99,7%; p<0,0001) with excellent screening parameters(sensitivity = 99,1%, specificity = 98,7%; p<0,0001)

AIRC-cCT automatically classified 130(37,4%) thoracic spines as abnormal, but only 18 (13,8%) of them were considered clinically relevant by the Expert Panel. 

Conclusions

Compared with the Expert Panel, AIRC-cCT detected and quantified emphysema and lung nodules with adequate accuracy, although with some false positive results. The software revealed great exactitude detecting thoracic aorta dilation. It was less precise for thoracic spine changes, with utmost false positive results.