Survival Time Prediction after Radiotherapy for Malignant Glioma Patients Based on Clinical and DVH Features Using Support Vector Machine

Abstract

Purpose/Objective(s): In the volumetric modulated arc therapy (VMAT), the planning configurations such as arc or collimator angles are selected on experientially, then the dose distribution was optimized iteratively. However, some cases may reselect the planning configurations when the resultant dose distribution was undesirable. To produce the adequate planning efficiently, the configurations should be optimized before dose distribution optimization. One of the most difficult subjects of it is the huge number of choices, it is not realistic to do for each patient. Design of experiments (DoE) is the one of techniques for optimization issues, it will identify the few factors which have large impact to results from large number of them, then obtained the optimized result by best combination of inputs. To obtain the adequate results with small number trials, DoE method was employed. Materials/Methods: CT data of the head and lung phantom were acquired using 16-slice CT (Light Speed 16-slice; GE Healthcare, Waukesha WI, USA). The target delineations were transformed from clinical CT using deformable image registration in mimicry of clinical situation, then modified it appropriately. The planning configurations such as arc angles, couch angles, or collimator angles were selected from two candidate values, respectively. To reduce the number of trials, orthogonal table was used which is the one of DoE tools to identify the main factor efficiently. By using the table, the trial number could reduce from 32 to 8 for each case. The dose volume histogram was calculated, then analyzed the best combination. Finally, dose distribution was calculated, then compared with it of conventional planning. Results: In the case of nasal cavity cancer (planed dose was 20 Gy), 3 fields were used, the best arc angles, couch angles, collimator angles, beam energies and size of beam window for field 1/2/3 were 135 -225 /135 -225 /70 -330 , 0 /0 /90 , 350 /80 /80 , 10X/10X/10X, and wide/wide/wide, respectively. The maximum dose of volume (Dmax) of left eye, right eye, spinal cord were 17.7 Gy, 7.0 Gy, 15.3 Gy in the conventional, and 15.8 Gy, 6.1 Gy, 14.8 Gy in the DoE, respectively. In the case of head and neck cancer (planed dose was 70 Gy), DoE was performed independently, Dmax of brain stem, spinal cord were 45.8 Gy, 43.6 Gy in the conventional, and 44.5Gy, 42.5 Gy in the DoE, respectively. The DoE configurations would have geometrical advantage to reduce organs at risk dose in these case. In the mediastinal lymph node metastasis (planed dose was 60 Gy), Dmax of esophagus, spinal cord, trachea were 24.4 Gy, 10.7 Gy, 8.7 Gy in the conventional, and 26.8 Gy, 10.4 Gy, 8.6 Gy in the DoE, respectively. We confirmed that the conventional configuration had been chosen the reasonable selection in this case. Conclusion: The optimization of beam configurations with DoE is expected to found the reasonable solution efficiently. These results indicate the effectiveness of this strategy for the clinical usage in the VMAT. Abstract 3664 Clinical featur...

Predictive performance of different NTCP techniques for radiation-induced esophagitis in NSCLC patients receiving proton radiotherapy

“…In the era of big data, the application of supervised machine learning in the treatment response modeling for radiotherapy has been rapidly growing. Support vector machine (SVM) 21 , a commonly used supervised machine learning algorithm, has shown promising accuracy in predicting lung radiation-induced pneumonitis 22 , local tumor control after stereotactic body radiation therapy for early-stage NSCLC patients 23 , and other clinical outcomes 24 , 25 . With the preset feature selection, the results of the comparison in performance between SVM and MLR were inconsistent 26 , 27 .…”