AUTHOR=Huang Peng , Shang Jiawen , Xu Yingjie , Hu Zhihui , Zhang Ke , Dai Jianrong , Yan Hui 

TITLE=Anomaly detection in radiotherapy plans using deep autoencoder networks

JOURNAL=Frontiers in Oncology

VOLUME=Volume 13 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2023.1142947

DOI=10.3389/fonc.2023.1142947

ISSN=2234-943X

ABSTRACT=Purpose: Treatment plans were used for patient under radiotherapy in clinic. Before execution they were checked for safety and quality by human experts. Few of them were identified with flaws and need further improvement. To automate this checking process, an unsupervised learning method based on autoencoder was proposed.
Methods: First, features were extracted from treatment plan by human experts. Then, these features were assembled and used for model learning. After network optimization, the reconstruction error between the predicted and target signals were obtained. Finally, the questionable plans were identified based on the value of reconstruction error. The large value of reconstruction error indicates the longer distance from the standard distribution of normal plans. A total of 576 treatment plans of breast cancer patients were used for the test. Among them, 19 were questionable plan identified by human experts. For evaluating the performance of autoencoder, it was compared with four baseline detection algorithms including local discrete factor (LOF), Density-based spatial clustering of applications with noise (HDBSCAN), one class of support vector machines (OC-SVM) and principal component analysis (PCA).
Results: The results showed that the autoencoder achieved the best performance than the other four baseline algorithms. The AUC value of the autoencoder was 0.9985, while the second one was 0.9535 (LOF). While maintaining 100% recall, the average accuracy and precision of the results by autoencoder were 0.9658 and 0.5143, respectively. While maintaining 100% recall, the average accuracy and precision of the results by LOF were by 0.8090 and 0.1472, respectively.
Conclusion: The autoencoder can effectively identify questionable plans from the large group of normal plans. There is no need to label data and prepare the training data for model learning. The autoencoder provided an effective way to carry out automate plan checking in radiotherapy.