[00657] Tomographic inverse problems and deep learning techniques
Session Time & Room : 3E (Aug.23, 17:40-19:20) @F401
Type : Proposal of Minisymposium
Abstract : Tomographic inverse problems involve the recovery of physical quantity $\left(\right.$shown as image$\left.\right)$ from indirect observations. These inverse problems are typically ill-posed and there have been numerous attempts to obtain the proper solution based on mathematical modeling. Deep learning has recently emerged as a powerful tool for solving the inverse problem, as it demonstrates the potential to handle uncertainty of the solution with large amounts of training data. In this mini-symposium, we will discuss mathematical and deep learning strategies for solving inverse problems related to imaging modalities such as computed tomography $\left(\right.$CT$\left.\right)$, magnetic resonance imaging $\left(\right.$MRI$\left.\right)$, and electrical impedance tomography $\left(\right.$EIT$\left.\right)$.
Organizer(s) : Kiwan Jeon, Hyoung Suk Park, Jin Keun Seo
[05264] Ill-posed Inverse problems in Low-dose Dental Cone-beam Computed Tomography
Format : Talk at Waseda University
Author(s) :
Hyoung Suk Park (National Institute for Mathematical Sciences)
Chang Min Hyun (Yonsei University)
Jin Keun Seo (Yonsei University)
Abstract : Dental cone-beam computed tomography (CBCT) has been increasingly being used in various dental fields such as implant/prosthetics, oral and maxillofacial surgery, and orthodontic treatment. It aims to provide high-resolution images with the lowest possible radiation dose at a low cost for equipment. However, this cost-competitive goal makes the inverse problem of dental CBCT more nonlinear and ill-posed. In this presentation, we describe the mathematical structure of an ill-posed nonlinear inverse problem of low-dose dental CBCT and explain the advantages and limitations of the deep learning-based approach for CBCT image reconstruction compared to conventional regularization methods.
[05265] Machine learning for automatic signal quality assessment in multi-channel electrical impedance-based hemodynamic monitoring
Format : Talk at Waseda University
Author(s) :
Chang Min Hyun (Yonsei University)
Abstract : Owing to recent advances in thoracic electrical impedance tomography, a patient's hemodynamic function can be noninvasively and continuously estimated in real-time by surveilling a cardiac volume signal associated with stroke volume and cardiac output. In clinical applications, however, a cardiac volume signal is often of low quality, mainly because of the patient's deliberate movements or inevitable motions during clinical interventions. This talk deals with developing a signal quality indexing method that assesses the influence of motion artifacts on transient cardiac volume signals. We apply divergent machine-learning methods from discriminative-model to manifold learning. The use of machine-learning could be suitable for our real-time monitoring application that requires fast inference and automation as well as high accuracy. In the clinical environment, the proposed method can be utilized to provide immediate warnings so that clinicians can minimize confusion regarding patients' conditions, reduce clinical resource utilization, and improve the confidence level of the monitoring system. Numerous experiments using actual EIT data validate the capability of cardiac volume signals degraded by motion artifacts to be accurately and automatically assessed in real-time by machine learning.
[05332] Application of AI-based Medical Diagnosis: focusing on tomography
Format : Talk at Waseda University
Author(s) :
Soomin Jeon (Dong-A University)
Abstract : In this talk, we will see the various applications of artificial intelligence (AI) based medical diagnosis, focusing on tomography. From data set configuration for AI medical image research using tomography such as PET, X-ray, and X-ray CT, we look at the results of research on training algorithms. We also cover research related to the latest diagnostic technologies such as Alzheimer's disease.
[05324] Data-driven reconstruction in X-ray CT with provable convergence guarantees
Author(s) :
Jürgen Frikel (OTH Regensburg)
Simon Göppel (University of Innsbruck)
Markus Haltmeier (University of Innsbruck)
Abstract : Computed tomography (CT) is a widely used imaging technique in various fields including medicine, engineering, and materials science. CT image reconstruction is an ill-posed problem, which means that small errors in the measurements can lead to significant errors in the reconstruction. In order to stabilize the image reconstruction, the use of regularization techniques is crucial. Classical regularization methods (e.g., variational regularization) are known to provide convergent methods, but the resulting reconstructions may not be optimal. On the other hand, regularization strategies based on machine learning methods have been shown to produce better reconstructions, but often lack theoretical convergence guarantees. To address these challenges, we propose a convergent data-driven reconstruction method for x-ray tomography. Our framework consists of two steps, where in a classical regularization is used in the first step and a deep learning approach is used in the second step. Both steps are coupled by integrating data proximity into a network architecture. This integration allows us to leverage the strengths of both approaches while addressing their limitations. Our experimental results demonstrate the effectiveness of our approach in improving the accuracy of CT image reconstruction, while providing convergence guarantees.
