Deep-learning-based hepatic fat assessment
The present disclosure relates to a method. The method includes accessing automatically segmented liver data and automatically segmented spleen data from a patient. The automatically segmented liver data is used to determine a liver attenuation and the automatically segmented spleen data is used to determine a spleen attenuation. A liver-to-spleen attenuation ratio is determined from the liver attenuation and the spleen attenuation. A hepatic steatosis determination is made from the liver-to-spleen attenuation ratio.
1 . A method, comprising:
accessing automatically segmented liver data and automatically segmented spleen data from a patient, wherein the automatically segmented liver data comprises a region of interest from a digitized image of a liver and the automatically segmented spleen data comprises a region of interest from a digitized image of a spleen;
utilizing the automatically segmented liver data to determine a liver attenuation;
utilizing the automatically segmented spleen data to determine a spleen attenuation;
determining a liver-to-spleen attenuation ratio from the liver attenuation and the spleen attenuation; and
making a hepatic steatosis determination from the liver-to-spleen attenuation ratio.
2 . The method of claim 1 , wherein the spleen attenuation and the liver attenuation are determined by a slice-based estimation method configured to identify a slice of the digitized image of the liver with a maximum liver area or a slice of the digitized image of the spleen with a maximum spleen area and to obtain a mean attenuation across an entire liver and an entire spleen within the slice.
3 . The method of claim 1 , wherein the spleen attenuation and the liver attenuation are determined by a volume-based estimation configured to obtain a mean attenuation across an entire liver and an entire spleen over a three-dimensional image volume.
4 . The method of claim 1 , wherein the liver-to-spleen attenuation ratio is obtained by:
measuring a first mean value of Hounsfield units over an entirety of a segmented liver identified by the automatically segmented liver data;
measuring a second mean value of Hounsfield units over an entirety of a segmented spleen identified by the automatically segmented spleen data; and
dividing the first mean value of Hounsfield units by the second mean value of Hounsfield units to obtain the liver-to-spleen attenuation ratio.
5 . The method of claim 1 , further comprising:
accessing one or more digitized images comprising a liver and a spleen;
operating one or more deep learning models on the one or more digitized images to segment the liver and generate the automatically segmented liver data;
operating the one or more deep learning models on the one or more digitized images to segment the spleen and generate the automatically segmented spleen data; and
storing the automatically segmented liver data and the automatically segmented spleen data in an electronic memory.
6 . The method of claim 5 , wherein the one or more digitized images comprise low-dose non-contrast computed tomography (CT) images.
7 . The method of claim 5 , wherein the one or more deep learning models comprise a 3D residual-UNet architecture from nnUnet framework.
8 . The method of claim 5 , further comprising:
training the one or more deep learning models on a plurality of computed tomography (CT) images.
9 . The method of claim 8 , wherein one or more of the plurality of CT images used in training do not include an entire liver.
10 . The method of claim 1 , further comprising:
determining a risk assessment metric based on the hepatic steatosis determination, wherein the risk assessment metric corresponds to a severity of COVID symptoms that the patient is expected to experience due to a COVID infection; and
assigning a care level to the patient based on the risk assessment metric.
11 . A non-transitory computer-readable medium storing computer-executable instructions that, when executed, cause a processor to perform operations, comprising:
operating one or more deep learning models on one or more computed tomography (CT) images comprising a liver and a spleen, wherein the one or more deep learning models are configured to segment the liver and generate automatically segmented liver data and to segment the spleen and generate automatically segmented spleen data;
measuring a liver attenuation from the automatically segmented liver data;
measuring a spleen attenuation from the automatically segmented spleen data;
determining a liver-to-spleen attenuation ratio from the liver attenuation and the spleen attenuation; and
generating a hepatic steatosis determination from the liver-to-spleen attenuation ratio.
12 . The non-transitory computer-readable medium of claim 11 , wherein the one or more deep learning models comprise a single deep learning model.
13 . The non-transitory computer-readable medium of claim 11 , wherein the operations further comprise:
determining one or more risk assessment metrics based on the hepatic steatosis determination.
14 . The non-transitory computer-readable medium of claim 13 , wherein the one or more risk assessment metrics comprise one or more of a COVID severity metric and a cardiovascular disease metric.
15 . The non-transitory computer-readable medium of claim 11 , wherein the one or more CT images comprise low-dose non-contrast computed tomography images.
16 . The non-transitory computer-readable medium of claim 11 , wherein the spleen attenuation and the liver attenuation are determined by a slice-based estimation method configured to identify a slice of a digitized image with a maximum area and obtain a mean attenuation across an entire liver and an entire spleen within the slice.
17 . The non-transitory computer-readable medium of claim 11 , wherein the spleen attenuation and the liver attenuation are determined by a volume-based estimation configured to obtain a mean attenuation across an entire liver and an entire spleen over a three-dimensional image volume.
18 . An apparatus, comprising:
one or more deep learning models configured to operate upon one or more digitized images that include a liver and a spleen to generate automatically segmented liver data and automatically segmented spleen data;
an attenuation calculation tool configured to utilize the automatically segmented liver data to measure a liver attenuation, to utilize the automatically segmented spleen data to measure a spleen attenuation, and to determine a liver-to-spleen attenuation ratio from the liver attenuation and the spleen attenuation; and
a hepatic steatosis calculation tool configured to generate a hepatic steatosis (HS) determination by comparing the liver-to-spleen attenuation ratio to an hepatic steatosis threshold.
19 . The apparatus of claim 18 , wherein the one or more deep learning models are configured to generate one or more binary masks that include the automatically segmented liver data identifying a segmented liver and that include the automatically segmented spleen data identifying a segmented spleen.
20 . The apparatus of claim 19 , wherein the liver-to-spleen attenuation ratio is obtained by:
measuring the liver attenuation as a first mean value of Hounsfield units over an entirety of the segmented liver;
measuring the spleen attenuation as a second mean value of Hounsfield units over an entirety of the segmented spleen; and
dividing the first mean value of Hounsfield units by the second mean value of Hounsfield units to obtain the liver-to-spleen attenuation ratio.