Deep-learning adds PCCT-like chest imaging quality to existing scanners

Sunday, November 29 | 9:00 a.m.- 10:00 a.m. | S1-SSCH01-4 | Room E451A

A deep-learning model called SUMI shows promise for "distilling [photon-counting CT] PCCT image-quality advantages into routine chest CT." 

Photon-counting CT (PCCT) offers higher spatial resolution and lower noise than conventional energy-integrating CT (EICT), but PCCT scanners aren't necessarily available for routine use, presenter Chuntung Zhuang of Johns Hopkins University in Baltimore and colleagues noted. To address the problem, the group crafted an AI model that incorporates PCCT image quality into routine CT imaging without paired acquisitions -- which could increase patient access to PCCT-like image quality without installing new hardware.

Zhuang's team first trained a continual autoencoder (i.e., an unsupervised artificial neural network) on 1,046 PCCT scans plus 405,379 EICT scans from 145 hospitals; it then trained SUMI using the 1,046 PCCT scans classified into four standard categories of image quality loss (sparse-view, low-dose, conventional, and mixed degradations).

Attend this session to find out how SUMI performed when it came to peak signal-to-noise ratio structural similarity index measure and how its use could improve the quality of chest CT.

Page 1 of 2
Next Page