International Journal of Hematology and Oncology
2025, Vol 35, Num 1 Page(s): 027-040
| Back | Table of Contents | PDF | Mail to Author | |
Radiomic-Assistant Response Prediction to Stereotactic Body Radiotherapy in Early Stage Lung Cancer
Melek YAKAR1, Durmus ETIZ1, Eyyup GULBANDILAR2, Kerem DURUER1, Ergin ERDEN1
1EskiSehir Osmangazi University, Faculty of Medicine, Department of Radiation Oncology
2EskiSehir Osmangazi University, Faculty of Engineering, Department of Computer Engineering
Keywords: Early-stage lung cancer, Stereotactic body radiotherapy, Radiomics, treatment response prediction, Artificial intelligence
The aim of this study was to predict SBRT response in patients with early-stage lung cancer who underwent SBRT using 4-dimensional computed tomography (4DCT) radiomics. 44 cases diagnosed with early-stage lung cancer and treated with SBRT between 2020-2024 were included in the study. The radiomic features of the patients were obtained from the planning 4DCT with the Lifex program. The LASSO method was used to determine important variables. The SMOTE method was used to create a balanced data set. SBRT response estimate (complete response/partial response/stable response) was created using artificial intelligence methods using important variables. Median BED10 was 100 (min: 72, max: 132) Gy. SBRT scheme was applied as 8-12.5 Gy x 4-6 fr. Median PFS and OS after SBRT were 15 and 20 months at median 20-month follow-up. SBRT response assessment was performed using RECIST criteria. Complete, partial and stable response rates among patients were 36.4%, 36.4% and 27.3%, respectively. 7 of 55 radiomic features obtained with Lifex program were determined as significant variables with LASSO method. Prediction models were created with 5 different artificial intelligence algorithms using 7 significant variables. When the test groups are examined, SBRT response prediction was performed with 71%, 78%, 64%, 92% and 72% accuracy rates using MLPNN-1, MLPNN-2, ANFIS-1, ANFIS-2 and MLPC algorithms, respectively. Radiomics are easy to obtain, non-invasive and contains patient-specific information. However, the imaging method, segmentation differences between users, obtaining Radiomics and creating prediction algorithms are quite heterogeneous, and standardization should be provided with multi-center studies with more patients. Radiomics can be a potential biomarker in SBRT response prediction when these steps are standardized. In the current study, the highest accuracy rate was created with the ANFIS-2 algorithm and studies with more patients are needed.
Melek YAKAR1, Durmus ETIZ1, Eyyup GULBANDILAR2, Kerem DURUER1, Ergin ERDEN1
1EskiSehir Osmangazi University, Faculty of Medicine, Department of Radiation Oncology
2EskiSehir Osmangazi University, Faculty of Engineering, Department of Computer Engineering
Keywords: Early-stage lung cancer, Stereotactic body radiotherapy, Radiomics, treatment response prediction, Artificial intelligence
The aim of this study was to predict SBRT response in patients with early-stage lung cancer who underwent SBRT using 4-dimensional computed tomography (4DCT) radiomics. 44 cases diagnosed with early-stage lung cancer and treated with SBRT between 2020-2024 were included in the study. The radiomic features of the patients were obtained from the planning 4DCT with the Lifex program. The LASSO method was used to determine important variables. The SMOTE method was used to create a balanced data set. SBRT response estimate (complete response/partial response/stable response) was created using artificial intelligence methods using important variables. Median BED10 was 100 (min: 72, max: 132) Gy. SBRT scheme was applied as 8-12.5 Gy x 4-6 fr. Median PFS and OS after SBRT were 15 and 20 months at median 20-month follow-up. SBRT response assessment was performed using RECIST criteria. Complete, partial and stable response rates among patients were 36.4%, 36.4% and 27.3%, respectively. 7 of 55 radiomic features obtained with Lifex program were determined as significant variables with LASSO method. Prediction models were created with 5 different artificial intelligence algorithms using 7 significant variables. When the test groups are examined, SBRT response prediction was performed with 71%, 78%, 64%, 92% and 72% accuracy rates using MLPNN-1, MLPNN-2, ANFIS-1, ANFIS-2 and MLPC algorithms, respectively. Radiomics are easy to obtain, non-invasive and contains patient-specific information. However, the imaging method, segmentation differences between users, obtaining Radiomics and creating prediction algorithms are quite heterogeneous, and standardization should be provided with multi-center studies with more patients. Radiomics can be a potential biomarker in SBRT response prediction when these steps are standardized. In the current study, the highest accuracy rate was created with the ANFIS-2 algorithm and studies with more patients are needed.
| Back | Table of Contents | PDF | Mail to Author | |