| Major accidents can happen during radiotherapy, with extremely severe consequences to both patients and clinical professionals. A primary cause of such accidents is a mistake in the patient's radiotherapy treatment chart. By applying data mining techniques to radiotherapy treatment planning, we developed an automatic Chart Checker to identify major errors in treatment charts. To determine what constitutes an error, we used computer-aided clustering of error-free charts to first distinguish common and acceptable values of the treatment parameters. Then, a new treatment chart is compared to the clustering results and if it contains values that largely deviate from the norm, it is marked as having errors and considered an "outlier". The errors are brought to the attention of the human chart checkers so that they can be corrected.;We applied an improved version of the K-Means clustering algorithm to a simple prostate cancer radiotherapy treatment model. 1000 treatment plans were used to build the clusters while another 650 plans were used to test our outlier detection method. It was found that there are eight distinct clusters of patients. In our model, we focused on two main sets of treatment parameters: beam energies and monitor units. Each set consists of four features that correspond to the direction in which the beam is administered: Anterior, Posterior, Right, and Left. By introducing +/-100% error to the monitor unit features, the detection rate is about 100%. At +/-50% error, the detection rate is about 80%. The false positive rate is about 10%. When purposely changing the beam energy to a value different from that in the treatment plan, the detection rate is 100% for posterior, right lateral, and left lateral fields, and about 77% for the anterior field. The results are very promising and our outlier detection methodology has the potential of identifying major errors in future radiotherapy treatment charts, thereby preventing devastating accidents. |
