TABLE 2.

Net Reclassification of Patients with CAD-DL (Explainable Deep Learning Model) Compared With the Same Readers Without CAD-DL

Reclassified 5-point scale likelihood with CAD-DLReclassified likelihood (%)
5-point scale likelihood without CAD-DLNormalProbably normalEquivocalProbably abnormalDefinitely abnormalIncreased likelihoodDecreased likelihoodNet correctly reclassified
Obstructive CAD (n = 360)47 (13.1%)25 (6.9%)6.1% (95% CI 1.4%–10.3%)
 Normal224200
 Probably normal4341033
 Equivocal0610140
 Probably abnormal0142411
 Definitely abnormal00010198
No Obstructive CAD (n = 360)40 (11.1%)80 (22.2%)11.1% (95% CI 4.8%–16.8%)
 Normal6411000
 Probably normal18841152
 Equivocal1131161
 Probably abnormal0915294
 Definitely abnormal1261552Overall NRI + 17.2% (95% CI 9.2%–24.4%, P < 0.001)
  • Boldface text indicates studies were reclassified in the correct direction and italicized text indicates studies were reclassified in the incorrect direction. A color version of Table 2 appears in the online supplemental materials.

  • NRI = net reclassification index.