Browse Prior Art Database

DEEP LEARNING BASED ATTENUATION CORRECTION FOR TIME-OF -FLIGHT POSITION EMISSION TOMOGRAPHY

IP.com Disclosure Number: IPCOM000249029D
Publication Date: 2017-Jan-27
Document File: 4 page(s) / 229K

Publishing Venue

The IP.com Prior Art Database

Abstract

A deep learning based attenuation correction technique for TOF-PET image reconstruction is proposed. The proposed technique allows direct estimation of attenuation correction map from TOF non-attenuation-corrected (TOF-NAC) images using the deep learning techniques. Deep learning based classifications algorithms are trained on PET TOF-NAC and computed tomography (CT) images using the large amount of PET/CT clinical data to identify the major component of the human body such as soft tissue, bone, and air in the TOF-NAC images. Such training enables construction of an image for attenuation correction. Advantageously, the proposed technique eliminates dependency on other image modalities.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 51% of the total text.

DEEP LEARNING BASED ATTENUATION CORRECTION FOR TIME-OF -FLIGHT POSITION EMISSION TOMOGRAPHY

BACKGROUND

 

The present disclosure relates generally to time-of-flight position emission tomography (TOF-PET) imaging system and more particularly to a deep learning based attenuation correction technique for TOF-PET image reconstruction.

Attenuation correction is a critical step for positron emission tomography (PET) image reconstruction. For whole-body PET, there are currently three commonly used techniques to obtain an attenuation map. The first technique is to acquire a transmission scan with external radionuclide sources. For work flow reasons, this scan is typically acquired in a few minutes and segmented to differentiate mainly the background, lungs, and soft tissue. The second technique is to combine PET and computed tomography (CT) into a PET/CT scanner, which uses spatially aligned CT data after a bilinear transformation to convert from CT Hounsfield units to attenuation factors at 511 keV. The third technique is to segment MR images into four classes (background, lungs, fat, soft tissue, for example) or five classes (bone as the additional class) and assign corresponding attenuation values to the classes.

Another technique to obtain attenuation map is simultaneous reconstruction of activity and attenuation for time-of-flight PET. This technique does not depend on any other imaging modality in principle. However, it benefits from additional information to reduce cross-talk artifact and improve convergence.   

It would be desirable to have an improved attenuation correction technique for PET image reconstruction that is not dependent on other imaging modalities.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 depicts working of the deep learning based attenuation technique. The TOF-PET data are first reconstructed into a 3D image volume without attenuation and scatter corrections (TOF-NAC images). Corresponding CT images are classified into label maps.

Figure 2 depicts training of the deep learning based network using the TOF-NAC images and label maps from CT images.

Figure 3 depicts use of the attenuation correction map as the initial attenuation map in  joint estimation of PET activity and attenuation.

DETAILED DESCRIPTION

A deep learning based attenuation correction technique for TOF-PET image reconstruction is proposed. The proposed technique allows direct estimation of attenuation correction map from TOF non-attenuation-corrected (TOF-NAC) images using the deep learning techniques. Deep learning based classifications algorithms are trained on PET TOF-NAC and computed tomography (CT) images using the large amount of PET/CT clinical data to identify the major component of the human body such as soft tissue, bone, and air in the TOF-NAC images. Such training enables construction of an image for attenuation correction without the images from other modalities.

Since quality of the TOF-NAC images has improved significantly in recent years with the...