A two‐dimensional feasibility study of deep learning‐based feature detection and characterization directly from CT sinograms

Abstract

Machine Learning, especially deep learning, has been used in typical x‐ray computed tomography (CT) applications, including image reconstruction, image enhancement, image domain feature detection and image domain feature characterization. To our knowledge, this is the first study on machine learning for feature detection and analysis directly based on CT projection data. Specifically, we present neural network methods for blood vessel detection and characterization in the sinogram domain avoiding any partial volume, beam hardening, or motion artifacts introduced during reconstruction. First, we estimate sinogram domain vessel maps using a residual encoder–decoder convolutional neural network (REDCNN). Next, we estimate the vessel centerline and we extract the vessel‐only sinogram from the original sinogram, eliminating any background information. Finally, we use a fully connected neural network to estimate the vessel lumen cross‐sectional area from the vessel‐only sinogram. We trained and tested the proposed methods using CatSim simulations, real CT measurements of vessel phantoms, and clinical data from the NIH CT image database. We achieved encouraging initial results showing the feasibility of CT analysis in the sinogram domain. In principle, sinogram domain analysis should be possible for many other and more complicated clinical CT analysis tasks. Further studies are needed for this sinogram domain analysis approach to become practical for clinical applications.

Making Radiomics More Reproducible across Scanner and Imaging Protocol Variations: A Review of Harmonization Methods

“…Most machine learning (ML) and DL algorithms have been used on reconstructed images in the existing medical imaging workflow. Instead, the abilities of ML and DL could be leveraged to process the underlying raw sensor-level data to access its hidden nuances [ 53 , 54 , 55 ]. A study conducted by Lee et al [ 56 ] investigated the performance of a CNN for classifying raw CT data in the sinogram-space to identify the body region and detect intracranial hemorrhage.…”

CoIL: Coordinate-based Internal Learning for Imaging Inverse Problems

“…GAN has been employed to synthesize missing measurements that are corrupted by the metal artifacts [70][71][72][73]. The effectiveness of these approaches has been shown in different imaging modalities [74][75][76]. Nevertheless, most deep synthesis methods require a dataset of fully-sampled measurements for supervised training.…”

In the Era of Deep Learning, Why Reconstruct an Image at All?

“…In these approaches, the whole image can be fed as input into the algorithm (ie, manual segmentation is not necessary), and the optimal features are learned during the training process. These methods are making progress in image segmentation (eg, tumor volume quantification), regression tasks (eg, bone age prediction using hand radiographs of children), registration (eg, longitudinal assessment of patient status), classification (eg, classifying tumors as benign versus malignant), and detection tasks (eg, localizing pneumonia) [4,6,7]. Although the ML and DL methods have their drawbacks, including the concern that they are "black boxes," lack generalizability and interpretability, and pose a risk for "brittleness" (ie, lacking robustness against changing inputs or conditions), these concerns may too be addressed in part through direct access to sensor-level data.…”