TY - JOUR ID - 13114 TI - Exploratory analysis of using supervised machine learning in [18F] FDG PET/CT images to predict treatment response in patients with metastatic and recurrent Brest tumors JO - Iranian Journal of Medical Physics JA - IJMP LA - en SN - AU - Nejabat, M. AU - Papp, L. AU - Monschein, L. AU - Hacker, M. AU - Beyer, T. AU - Leisser, A. AU - Haug, A.R. AD - Department of Biomedical Imaging and Image-guided Therapy, Division of Nuclear Medicine, Medical University of Vienna, Austria AD - Department of Biomedical Imaging and Image-guided Therapy, Division of Nuclear Medicine, Medical University of Vienna, Austria Center of Medical Physics and Biomedical Engineering, QIMP Group, Medical University of Vienna Y1 - 2018 PY - 2018 VL - 15 IS - Special Issue-12th. Iranian Congress of Medical Physics SP - 425 EP - 425 KW - Machine Learning KW - FDG PET/CT images KW - Brest tumors DO - 10.22038/ijmp.2018.13114 N2 - Aim: Despite grate progress in treatments, breast cancer is still the most common invasive cancer and the most cause of cancer related death in women. Treatment could be improved and perhaps standardized if more reliable markers for tumour progression and poor prognosis could be developed. The aim of this study was to evaluate whether patient-based machine learning (ML) driven analysis of 2-deoxy-2-(18F) fluoro-D-glucose PET/CT ([18F] FDG-PET/CT) is feasible to predict for treatment response and overall survival (OS) in patients with ENT tumours. Materials and methods: In total 136 patients with the diagnosis of metastatic and recurrent breast cancer (ductal/Lobular), who had a positive [18F] FDG- PET/CT scan between 12/2008 and 12/2015 were included in this analysis. Up to five malignant lesions were delineated on the PET images using semi-automatic VOIs, which were summed up to one total tumour volume, followed by feature extraction. Clinical data such as age, tumour grade, OS, course of treatment, response and P53, HER2, ER (oestrogen receptor) and PR (progesterone receptor) status were collected. ML approaches were utilized to identify relevant textural features on PET/CT and patient features and their relative weights for survival and response prediction. The established models were validated in a Monte Carlo (MC) cross-validation scheme, as presented in Papp et al. The individual datasets for these ML executions was selected from the given MC subset by bootstrapping.   Results: Median OS was 20.0 months (range: 0-89 mo). 46 patients received chemotherapy (8 patients received surgical resection after chemotherpy), 34 resections, 37 radiations and 23 hormonotherapy; response rates were 21/46, 18/35, 15/37 and 9/23 for the four treatment groups respectively. A treatment-based subgroup analysis yielded the best results with sensitivity (SNS) of 0.76, specificity (SPC) of 0.68 and an area under the curve (AUC) of 0.72 predicting for response and SNS of 0.7, SPC of 0.8 and AUC of 0.8 predicting for tumour grade after chemotherapy. For the whole cohort prediction of OS, response and grade showed values of 0.8, 0.6 and 0.6, 0.5, 0.6 and 0.7, 0.6, 0.6 and 0.7 for SNS, SPC and AUC respectively. Conclusion: These results demonstrate that textural and joint fusion features from PET-CT obtained by supervised ML are a valuable option for predicting OS, response and tumour grade in breast tumours. UR - https://ijmp.mums.ac.ir/article_13114.html L1 - ER -