Skip to main content
. 2023 Mar 10;9(3):182–194. doi: 10.1002/cjp2.314

Table 2.

CNN, SVM, and RF model analyses summarized into case‐based results

(A)
Prediction
RG0 RG1 RG2 RG3 Total
Truth RG0 9 0 0 1 10
RG1 0 34 1 9 44
RG2 1 0 19 1 21
RG3 1 0 0 27 28
Total 11 34 20 38 103
(B)
Prediction
RG0 RG1 RG2 RG3 Total
Truth RG0 5 3 1 1 10
RG1 0 41 0 3 44
RG2 1 5 13 2 21
RG3 0 2 0 26 28
Total 6 51 14 32 103
(C)
Prediction
RG0 RG1 RG2 RG3 Total
Truth RG0 6 0 2 2 10
RG1 0 43 0 1 44
RG2 0 4 14 3 21
RG3 0 3 0 25 28
Total 6 50 16 31 103
(D)
Prediction
RG0 RG1 RG2 RG3 Total
Truth RG0 9 1 0 0 10
RG1 0 42 2 0 44
RG2 0 0 21 0 21
RG3 0 0 0 28 28
Total 9 43 23 28 103

(A) CNN model analysis result of test cases (accuracy: 86.4%; 95% CI: 78.3–92.4%; κ = 0.81). (B) SVM model analysis (only RG information was used) result of test cases (accuracy: 82.5%; 95% CI: 73.8–89.3%; κ = 0.74). (C) SVM model analysis (RG and subtype information was used) result of test cases (accuracy: 85.4%; 95% CI: 77.1–91.6%; κ = 0.78). (D) RF model analysis (RG and subtype information was used) result of test cases (accuracy: 97.1%; 95% CI: 91.7–99.4%; κ = 0.96).