Table 3 Comparison of categorization performance between the CNN and radiologists.

From: Diagnosis of thyroid micronodules on ultrasound using a deep convolutional neural network

Test

Linear trend χ2 testa

LR χ2 testa

AICb

CNN TIRADS

20.3

20.9

264.8

KSThR TIRADS

7.0

6.3

279.4

  1. CNN convolutional neural network, LR likelihood ratio, AIC Akaike information criterion, KSThR Korean Society of Thyroid Radiology, TIRADS Thyroid Imaging Reporting and Data System.
  2. aHigher values suggest better monotonicity of gradient and heterogeneity. bLower values suggest a more parsimonious model.
Back to article page