Related Experiment Video
Updated: Mar 28, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Use of artificial neural networks to predict the probability of developing new cerebral metastases after radiosurgery
Parisa Azimi1, Sohrab Shahzadi2, Sohrab Sadeghi2
1Functional Neurosurgery Research Center, Shahid Beheshti University of Medical Sciences, Tehran, Iran - parisa.azimi@gmail.com.
Background:
The present study aimed to predict the probability of developing new cerebral metastases after Gamma Knife radiosurgery (GKR) alone in patients with 1-3 brain metastases by artificial neural network (ANN) model.
Methods:
AAN and other methods were used. The AAN has been investigated in comparison of other models of analysis, including the logistic regression (LR) and the recently introduced Huttenlocher method. Patients were divided into a distant brain failure (DBF) or a distant brain control (DBC) within 6 months of follow-up. Eleven factors were determined as the input parameters for the established ANN and were trained to predict of DBF. Sensitivity analysis was performed. The ROC curve, accuracy rate, and Hosmer-Lemeshow (H-L) tests were used to assess the three different prediction methods.
Results:
A total of 192 patients were categorized into the training (N.=96), testing (N.=48), and validation (N.=48) data sets in ANN. DBC rate was 60.4; based on the Huttenlocher classification method, it was 40.7% in Group I, 49.2% in Group II, and 88.1% in Group III. The number of cerebral metastases, primary tumor type, extra-cerebral metastases and recursive partitioning analysis (RPA) were the most important variables indicated by the ANN analysis. Compared to the other two methods, ANN model was associated to superior results: accuracy rate, 95.3%; H-L statistic, 40.9%; and AUC, 0. 0.88%.
Conclusions:
The ANNs can be used to effectively help for predicting DBF in patients with 1-3 brain metastasis treated with GKR alone.
