RESUMEN
BACKGROUND: Radiomics shows promising results in supporting the clinical decision process, and much effort has been put into its standardization, thus leading to the Imaging Biomarker Standardization Initiative (IBSI), that established how radiomics features should be computed. However, radiomics still lacks standardization and many factors, such as segmentation methods, limit study reproducibility and robustness. AIM: We investigated the impact that three different segmentation methods (manual, thresholding and region growing) have on radiomics features extracted from 18F-PSMA-1007 Positron Emission Tomography (PET) images of 78 patients (43 Low Risk, 35 High Risk). Segmentation was repeated for each patient, thus leading to three datasets of segmentations. Then, feature extraction was performed for each dataset, and 1781 features (107 original, 930 Laplacian of Gaussian (LoG) features, 744 wavelet features) were extracted. Feature robustness and reproducibility were assessed through the intra class correlation coefficient (ICC) to measure agreement between the three segmentation methods. To assess the impact that the three methods had on machine learning models, feature selection was performed through a hybrid descriptive-inferential method, and selected features were given as input to three classifiers, K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), Random Forest (RF), AdaBoost and Neural Networks (NN), whose performance in discriminating between low-risk and high-risk patients have been validated through 30 times repeated five-fold cross validation. CONCLUSIONS: Our study showed that segmentation methods influence radiomics features and that Shape features were the least reproducible (average ICC: 0.27), while GLCM features the most reproducible. Moreover, feature reproducibility changed depending on segmentation type, resulting in 51.18% of LoG features exhibiting excellent reproducibility (range average ICC: 0.68-0.87) and 47.85% of wavelet features exhibiting poor reproducibility that varied between wavelet sub-bands (range average ICC: 0.34-0.80) and resulted in the LLL band showing the highest average ICC (0.80). Finally, model performance showed that region growing led to the highest accuracy (74.49%), improved sensitivity (84.38%) and AUC (79.20%) in contrast with manual segmentation.
RESUMEN
BACKGROUND/AIM: Nowadays, Machine Learning (ML) algorithms have demonstrated remarkable progress in image-recognition tasks and could be useful for the new concept of precision medicine in order to help physicians in the choice of therapeutic strategies for brain tumours. Previous data suggest that, in the central nervous system (CNS) tumours, amino acid PET may more accurately demarcate the active disease than paramagnetic enhanced MRI, which is currently the standard method of evaluation in brain tumours and helps in the assessment of disease grading, as a fundamental basis for proper clinical patient management. The aim of this study is to evaluate the feasibility of ML on 11[C]-MET PET/CT scan images and to propose a radiomics workflow using a machine-learning method to create a predictive model capable of discriminating between low-grade and high-grade CNS tumours. MATERIALS AND METHODS: In this retrospective study, fifty-six patients affected by a primary brain tumour who underwent 11[C]-MET PET/CT were selected from January 2016 to December 2019. Pathological examination was available in all patients to confirm the diagnosis and grading of disease. PET/CT acquisition was performed after 10 min from the administration of 11C-Methionine (401-610 MBq) for a time acquisition of 15 min. 11[C]-MET PET/CT images were acquired using two scanners (24 patients on a Siemens scan and 32 patients on a GE scan). Then, LIFEx software was used to delineate brain tumours using two different semi-automatic and user-independent segmentation approaches and to extract 44 radiomics features for each segmentation. A novel mixed descriptive-inferential sequential approach was used to identify a subset of relevant features that correlate with the grading of disease confirmed by pathological examination and clinical outcome. Finally, a machine learning model based on discriminant analysis was used in the evaluation of grading prediction (low grade CNS vs. high-grade CNS) of 11[C]-MET PET/CT. RESULTS: The proposed machine learning model based on (i) two semi-automatic and user-independent segmentation processes, (ii) an innovative feature selection and reduction process, and (iii) the discriminant analysis, showed good performance in the prediction of tumour grade when the volumetric segmentation was used for feature extraction. In this case, the proposed model obtained an accuracy of ~85% (AUC ~79%) in the subgroup of patients who underwent Siemens tomography scans, of 80.51% (AUC 65.73%) in patients who underwent GE tomography scans, and of 70.31% (AUC 64.13%) in the whole patients' dataset (Siemens and GE scans). CONCLUSIONS: This preliminary study on the use of an ML model demonstrated to be feasible and able to select radiomics features of 11[C]-MET PET with potential value in prediction of grading of disease. Further studies are needed to improve radiomics algorithms to personalize predictive and prognostic models and potentially support the medical decision process.