In proton beam therapy, the Bragg peak is the point where protons lose energy the fastest. This point is crucial for dose control, preserving healthy tissues, minimizing lateral scattering, and the success of treatment planning. However, accurately predicting the location of the Bragg peak is challenging due to the complex interactions of protons with tissues. This study proposes a machine learning (ML) approach to predict the exact location of the Bragg peak from phantom tissue proton beam therapy experiments. A dataset comprising the eight most commonly used biomaterials, which mimic human tissue in proton therapy procedures, has been curated for this study. Various ML models are benchmarked to find the most successful approach. ML model parameters are further optimized using a metaheuristic approach to achieve the highest prediction capability. In addition, feature contributions of each feature in the dataset are analyzed using an explainable artificial intelligence (XAI) technique. According to experimental results, Random Forest (RF) model that is optimized with Genetic Algorithm (GA) achieved 0.742 Correlation Coefficient (CC) value, 0.069 Mean Absolute Error (MAE) and 0.145 Root Mean Square Error (RMSE) outperforming other ML models. The proposed approach can track and predict the movement of the proton beam in real-time during treatment, enhancing treatment safety and contributing to the more effective management of the treatment process. This study is the first to predict exact Bragg curve peak locations from proton beam therapy experiments using ML approaches. The optimized ML model can provide higher precision in identifying the needed beam dosage for targeted tumor and improving treatment outcomes.
Birincil Dil | İngilizce |
---|---|
Konular | Bilgi Sistemleri (Diğer) |
Bölüm | Research Article |
Yazarlar | |
Yayımlanma Tarihi | |
Gönderilme Tarihi | 10 Ocak 2024 |
Kabul Tarihi | 5 Mart 2024 |
Yayımlandığı Sayı | Yıl 2024 Cilt: 66 Sayı: 2 |
Communications Faculty of Sciences University of Ankara Series A2-A3 Physical Sciences and Engineering
This work is licensed under a Creative Commons Attribution 4.0 International License.