Abstract:
The success of the intracranial aneurysm treatment by coil embolization depends on the size of coils inserted into an aneurysm. Since it is difficult to predict how coils distribute inside the aneurysm, radiologists often select coil dimension (the shape diameter (SD) and the length) by experience. The aim of this study is to design a system to model the selection pattern of the radiologist for the first three coils inserted into the aneurysm. Only the SD selection is modeled, since the radiologist often selects the longest available coils. Regression systems (RS), classification systems (CS) and hybrid systems (HS) were investigated. 87 training data were used to create the model. The efficiency of the three systems was measured using the leave-one-out cross validation (LOOCV) method. The LOOCV results indicates that the RS should be used for the first coil selection and the HS consisting of Bagging classifier and the RS should be used for the selection of the second and the third coils. According to the experiment on 13 validating dataset, the SD should be selected by CS if it had at least 10 training data; otherwise, it should be selected by RS. In most cases, the predicted SD was within 1mm of the SD used in the actual treatment. In one aneurysm, different radiologists may use coil with different SD, so the interactive system is implemented such that users can select SD within 1mm of the predicted SD. In vitro experiment was performed to ensure the effectiveness and the safety of the proposed interactive system.