RESUMO
PURPOSES: Radiomics is a quantitative method able to analyze a high-throughput extraction of minable imaging features. Herein, we aim to develop a CT angiography-based radiomics analysis and machine learning model for carotid plaques to discriminate vulnerable from no vulnerable plaques. MATERIALS AND METHODS: Thirty consecutive patients with carotid atherosclerosis were enrolled in this pilot study. At surgery, a binary classification of plaques was adopted ("hard" vs "soft"). Feature extraction was performed using the R software package Moddicom. Pairwise feature interdependencies were evaluated using the Spearman rank correlation coefficient. A univariate analysis was performed to assess the association between each feature and the plaque classification and chose top-ranked features. The feature predictive value was investigated using binary logistic regression. A stepwise backward elimination procedure was performed to minimize the Akaike information criterion (AIC). The final significant features were used to build the models for binary classification of carotid plaques, including logistic regression (LR), support vector machine (SVM), and classification and regression tree analysis (CART). All models were cross-validated using fivefold cross validation. Class-specific accuracy, precision, recall and F-measure evaluation metrics were used to quantify classifier output quality. RESULTS: A total of 230 radiomics features were extracted from each plaque. Pairwise Spearman correlation between features reported a high level of correlations, with more than 80% correlating with at least one other feature at |ρ|> 0.8. After a stepwise backward elimination procedure, the entropy and volume features were found to be the most significantly associated with the two plaque groups (p < 0.001), with AUCs of 0.92 and 0.96, respectively. The best performance was registered by the SVM classifier with the RBF kernel, with accuracy, precision, recall and F-score equal to 86.7, 92.9, 81.3 and 86.7%, respectively. The CART classification tree model for the entropy and volume features model achieved 86.7% well-classified plaques and an AUC of 0.987. CONCLUSION: This pilot study highlighted the potential of CTA-based radiomics and machine learning to discriminate plaque composition. This new approach has the potential to provide a reliable method to improve risk stratification in patients with carotid atherosclerosis.
Assuntos
Doenças das Artérias Carótidas , Placa Aterosclerótica , Algoritmos , Artérias Carótidas , Doenças das Artérias Carótidas/diagnóstico por imagem , Angiografia por Tomografia Computadorizada , Humanos , Projetos Piloto , Placa Aterosclerótica/diagnóstico por imagemRESUMO
Objective.Respiratory motion of lung tumours and adjacent structures is challenging for radiotherapy. Online MR-imaging cannot currently provide real-time volumetric information of the moving patient anatomy, therefore limiting precise dose delivery, delivered dose reconstruction, and downstream adaptation methods.Approach.We tailor a respiratory motion modelling framework towards an MR-Linac workflow to estimate the time-resolved 4D motion from real-time data. We develop a multi-slice acquisition scheme which acquires thick, overlapping 2D motion-slices in different locations and orientations, interleaved with 2D surrogate-slices from a fixed location. The framework fits a motion model directly to the input data without the need for sorting or binning to account for inter- and intra-cycle variation of the breathing motion. The framework alternates between model fitting and motion-compensated super-resolution image reconstruction to recover a high-quality motion-free image and a motion model. The fitted model can then estimate the 4D motion from 2D surrogate-slices. The framework is applied to four simulated anthropomorphic datasets and evaluated against known ground truth anatomy and motion. Clinical applicability is demonstrated by applying our framework to eight datasets acquired on an MR-Linac from four lung cancer patients.Main results.The framework accurately reconstructs high-quality motion-compensated 3D images with 2 mm3isotropic voxels. For the simulated case with the largest target motion, the motion model achieved a mean deformation field error of 1.13 mm. For the patient cases residual error registrations estimate the model error to be 1.07 mm (1.64 mm), 0.91 mm (1.32 mm), and 0.88 mm (1.33 mm) in superior-inferior, anterior-posterior, and left-right directions respectively for the building (application) data.Significance.The motion modelling framework estimates the patient motion with high accuracy and accurately reconstructs the anatomy. The image acquisition scheme can be flexibly integrated into an MR-Linac workflow whilst maintaining the capability of online motion-management strategies based on cine imaging such as target tracking and/or gating.
Assuntos
Neoplasias Pulmonares , Radioterapia Guiada por Imagem , Humanos , Neoplasias Pulmonares/diagnóstico por imagem , Neoplasias Pulmonares/radioterapia , Imageamento por Ressonância Magnética/métodos , Movimento (Física) , Imageamento Tridimensional , Respiração , Radioterapia Guiada por Imagem/métodosRESUMO
An MR-Linac can provide motion information of tumour and organs-at-risk before, during, and after beam delivery. However, MR imaging cannot provide real-time high-quality volumetric images which capture breath-to-breath variability of respiratory motion. Surrogate-driven motion models relate the motion of the internal anatomy to surrogate signals, thus can estimate the 3D internal motion from these signals. Internal surrogate signals based on patient anatomy can be extracted from 2D cine-MR images, which can be acquired on an MR-Linac during treatment, to build and drive motion models. In this paper we investigate different MRI-derived surrogate signals, including signals generated by applying principal component analysis to the image intensities, or control point displacements derived from deformable registration of the 2D cine-MR images. We assessed the suitability of the signals to build models that can estimate the motion of the internal anatomy, including sliding motion and breath-to-breath variability. We quantitatively evaluated the models by estimating the 2D motion in sagittal and coronal slices of 8 lung cancer patients, and comparing them to motion measurements obtained from image registration. For sagittal slices, using the first and second principal components on the control point displacements as surrogate signals resulted in the highest model accuracy, with a mean error over patients around 0.80 mm which was lower than the in-plane resolution. For coronal slices, all investigated signals except the skin signal produced mean errors over patients around 1 mm. These results demonstrate that surrogate signals derived from 2D cine-MR images, including those generated by applying principal component analysis to the image intensities or control point displacements, can accurately model the motion of the internal anatomy within a single sagittal or coronal slice. This implies the signals should also be suitable for modelling the 3D respiratory motion of the internal anatomy.