01-02-2015 | Original Article
Diagnostic performance of bone scintigraphy analyzed by three artificial neural network systems
Published in: Annals of Nuclear Medicine | Issue 2/2015
Login to get accessAbstract
Objective
The accuracy of bone scintigraphy analyzed by computer-assisted diagnosis (CAD) software involving multiple artificial neural network (ANN) systems has not been well established.
Methods
We conducted a retrospective study to examine the accuracy of bone scintigraphy analyzed by CAD software, BONENAVI® version 2 (BN2; FUJIFILM RI Pharma Co., Ltd.), in patients with suspected bone metastases. In 399 patients, bone metastases were analyzed by means of the BN2 focused on balance of sensitivity and specificity ANN system (BN2-B), focused on specificity-ANN system (BN2-Sp), and focused on sensitivity-ANN system (BN2-Sen). The ANN presented an output between 0 and 1 for each patient. A cutoff value of 0.5 was chosen to provide BN2 with the binary classification of “bone metastasis” or “no bone metastasis”. The area under the receiver operating characteristic curve (AUC) was used to evaluate diagnostic accuracy.
Results
A total of 18 % of male patients (36/196) and 12 % of female patients (24/203) had bone metastases. BN2-Sp and BN2-Sen were similar to BN2-B in the ability to identify patients who had bone metastases; the AUC values were 0.87 [95 % confidence interval (CI) 0.79–0.95], 0.92 (95 % CI 0.85–0.97), and 0.90 (95 % CI 0.83–0.97), respectively, in male patients. In female patients, the AUC values were 0.81 (95 % CI 0.71–0.91), 0.85 (95 % CI 0.78–0.93), and 0.81 (95 % CI 0.71–0.92), respectively. A total of 65.4 % of patients were classified as concordance of “bone metastases” (17.8 %) or “no bone metastases” (47.6 %), and 34.6 % were classified as mismatch. In the concordance group, BN2-B revealed an AUC of 0.94 (95 % CI 0.88–0.99), with a sensitivity of 94 % (95 % CI 79–98 %) and a specificity of 88 % (95 % CI 79–93 %) in 120 male patients and an AUC of 0.89 (95 % CI 0.78–1.00), with a sensitivity of 86 % (95 % CI 60–96 %) and a specificity of 85 % (95 % CI 78–90 %) in 141 female patients.
Conclusions
Bone scintigraphy analyzed by BN2-B accurately identifies the presence of bone metastases in patients with concordance using three ANN systems, comprising 65 % of the patients we studied.