US 11,657,504 B1
System and method for computationally efficient artificial intelligence based point-of-care ultrasound imaging healthcare support
Muhammad Moinuddin, Jeddah (SA); Ubaid M. Al-Saggaf, Jeddah (SA); Mohammed Jamal Abdulaal, Jeddah (SA); and Abdulrahman U. Alsaggaf, Jeddah (SA)
Assigned to King Abdulaziz University, Jeddah (SA)
Filed by King Abdulaziz University, Jeddah (SA)
Filed on Jun. 28, 2022, as Appl. No. 17/851,485.
Int. Cl. G06T 7/00 (2017.01); G06T 5/00 (2006.01)
CPC G06T 7/0012 (2013.01) [G06T 5/002 (2013.01); G06T 2207/10132 (2013.01); G06T 2207/20024 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30024 (2013.01)] 11 Claims
OG exemplary drawing
 
1. A reduced computation, computer-based method for management of point-of-care (POC) on-site ultrasound (ULS) imaging resources, comprising:
receiving at a digital computation resource of a POC site, from a hand-held ULS scanning device, a ULS tissue reflection sample data;
applying, by the digital computation resource of the POC site, a reduced computation ULS tissue reflection speckle noise (SN) physics model—blurring noise (BN) physics model based deep learning (DL) trained convolutional neural network (CNN) denoising processing to the ULS tissue reflection sample data, which outputs an estimated denoised ULS tissue reflection image data; and
displaying a visual rendering of the estimated denoised ULS tissue reflection image data, on a display resource of the POC site,
wherein:
the reduced computation ULS tissue reflection SN physics model—BN physics model based DL trained CNN denoising processing comprises a densely connected block of N multi-rate-multi-filter (MRMF) feature map generating processes, N being an integer, each MRMF feature map generating processing comprises receiving a respective ULS data input and generating, in response, a plurality of different spatial filter feature maps,
generating at least one of the spatial filter feature maps comprises convolving a respective spatial filter with the respective ULS data input, and feeding a result of the convolving to an activation processing that, in response, outputs the at least one spatial filter feature map,
the spatial filter comprises a weight,
the weight is an optimized weight, based on DL training using DL training data corresponding to the ULS tissue reflection SN physics model—BN physics model,
the activation processing comprises an activation parameter, and
the activation parameter is an optimized activation parameter, based on the DL training using the DL training data corresponding to the ULS tissue reflection SN physics model—BN physics model.