AI performance
The AI system achieved an overall accuracy of 79.8% (95% CI
77.0-82.6%) in correctly identifying each type of CNS malformation,
with a sensitivity of 78.4% (75.3-81.3%), specificity of 94.4%
(86.2-98.4%) and an AUC of 0.864 (0.833-0.895). The performance of CPC
identification was the best among all types of malformations detection,
with a sensitivity of 92.0% (74.0-99.0%), specificity of 99.9%
(99.3-100%)and AUC 0.959(0.905-1.000). Whereas, the performance of
Blake’s pouch cyst diagnosis was the lowest in terms of sensitivity of
42.9% (21.8- 66.0%), specificity of 99.6% (98.9-99.9%), and AUC
0.712 (0.604- 0.821). The diagnostic efficacy for the total and specific
types of anomalies identification were shown in Table 2.