Skip to main content
. Author manuscript; available in PMC: 2017 Sep 1.
Published in final edited form as: Artif Intell Med. 2016 Jul 27;72:42–55. doi: 10.1016/j.artmed.2016.07.001

Table 5.

The true positive (tp), false negative (fn), false potive (fp) and true negative (tn) rates of the DBN and Logistic Regression models. DBN A & B Whole Dataset: The average tp, fn, fp and tn rates of the 10 DBN models trained on 400 cases and evaluated on the remaining NLST dataset of 25, 446 cases. DBN A & B random sets: The average tp, fn, fp and tn rates of the 10 DBN models trained and tested on random balanced sets of 400 cases. Physicians Whole Dataset: The average tp, fn, fp and tn rates of the physicians classifications on the entire NLST dataset of 25, 446 cases. Physicians random sets: The average tp, fn, fp and tn rates of the physicians classifications on the random balanced sets of 400 cases. LR model: The full LR model without spiculation [41] evaluated on 5, 353 cases with nodule information of the NLST at baseline (t = 0). The parameters of the model used in this evaluation were adopted from [41]. LR-trained model: The average tp, fn, fp and tn rates of the logistic regression (LR) model, at baseline (t = 0), trained on 2, 663 cases and tested on 2, 690 cases. Dashes represent the cases for which no tp, fn, fp and tn rates were computed for the LR models.

First Screening Second Screening Third Screening
DBN A, whole dataset 92.6% (tp)
30.2% (fp)		 7.40% (fn)
69.8% (tn)		 87.3% (tp)
9.40% (fp)		 12.7% (fn)
90.6% (tn)		 83.9% (tp)
6.70% (fp)		 16.1% (fn)
93.3% (tn)
DBN A, random test sets 96.4% (tp)
65.6% (fp)		 3.60% (fn)
35.4% (tn)		 87.1% (tp)
17.0% (fp)		 12.9% (fn)
83.0% (tn)		 83.3% (tp)
9.60% (fp)		 16.7% (fn)
90.4% (tn)
DBN B, whole dataset 92.6% (tp)
30.2% (fp)		 7.40% (fn)
69.8% (tn)		 87.3% (tp)
9.40% (fp)		 12.7% (fn)
90.6% (tn)		 83.9% (tp)
6.9% (fp)		 16.1% (fn)
93.1% (tn)
DBN B, random test sets 92.7% (tp)
39.2% (fp)		 7.30% (fn)
60.8% (tn)		 87.1% (tp)
17.0% (fp)		 12.9% (fn)
83.0% (tn)		 83.3% (tp)
9.60% (fp)		 16.7% (fn)
90.4% (tn)
Physicians, whole dataset 89.7% (tp)
23.2% (fp)		 10.3% (fn)
76.8% (tn)		 93.7% (tp)
15.0% (fp)		 6.30% (fn)
85.0% (tn)		 90.1% (tp)
9.60% (fp)		 9.90% (fn)
90.4% (tn)
Physicians, random test sets 89.1% (tp)
31.5% (fp)		 10.9% (fn)
68.5% (tn)		 93.6% (tp)
20.7% (fp)		 6.40% (fn)
79.3% (tn)		 90.2% (tp)
12.8% (fp)		 9.80% (fn)
87.3% (tn)
Logistic Regression [41] 11.0% (tp)
0.50% (fp)		 89.0% (fn)
99.5% (tn)		 -
-		 -
-		 -
-		 -
-
Logistic Regression, retrained 16.3% (tp)
0.8% (fp)		 83.7% (fn)
99.2% (tn)		 -
-		 -
-		 -
-		 -
-