2022/03/11 by Arman Rahmim, Rahmim, Arman, Amirhosein Toosi +21 · 1 citation
Medicine · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Physical sciences #Inflammatory Biomarkers in Disease Prognosis #Lung Cancer Diagnosis and Treatment #Medical Physics (physics.med-ph) #Radiomics and Machine Learning in Medical Imaging
paper · pdf · doi:10.48550/arxiv.2203.06314
openalex publication_date 2022/03/11 · openalex created_date 2022/05/05 · openalex updated_date 2026/07/28
Radiomics features extract quantitative information from medical images,\ntowards the derivation of biomarkers for clinical tasks, such as diagnosis,\nprognosis, or treatment response assessment. Different image discretization\nparameters (e.g. bin number or size), convolutional filters, segmentation\nperturbation, or multi-modality fusion levels can be used to generate radiomics\nfeatures and ultimately signatures. Commonly, only one set of parameters is\nused; resulting in only one value or flavour for a given RF. We propose tensor\nradiomics (TR) where tensors of features calculated with multiple combinations\nof parameters (i.e. flavours) are utilized to optimize the construction of\nradiomics signatures. We present examples of TR as applied to PET/CT, MRI, and\nCT imaging invoking machine learning or deep learning solutions, and\nreproducibility analyses: (1) TR via varying bin sizes on CT images of lung\ncancer and PET-CT images of head & neck cancer (HNC) for overall survival\nprediction. A hybrid deep neural network, referred to as TR-Net, along with two\nML-based flavour fusion methods showed improved accuracy compared to regular\nrediomics features. (2) TR built from different segmentation perturbations and\ndifferent bin sizes for classification of late-stage lung cancer response to\nfirst-line immunotherapy using CT images. TR improved predicted patient\nresponses. (3) TR via multi-flavour generated radiomics features in MR imaging\nshowed improved reproducibility when compared to many single-flavour features.\n(4) TR via multiple PET/CT fusions in HNC. Flavours were built from different\nfusions using methods, such as Laplacian pyramids and wavelet transforms. TR\nimproved overall survival prediction. Our results suggest that the proposed TR\nparadigm has the potential to improve performance capabilities in different\nmedical imaging tasks.\n