Devices and methods for detecting pulmonary function based on low-dose CT images

1 . A method of determining a pulmonary function based on low-dose computed tomography (CT) chest images, comprising:

receiving a plurality of chest images, the plurality of chest image generated by a low-dose CT method;

determining a plurality of regions of interest (ROIs) within each chest image through applying the plurality of chest images to an image processing model, the plurality of ROIs corresponding to five lung lobes;

determining T ij descriptors for the j-th ROI of the i-th chest image, each descriptor associated with a radiodensity value of one or more pixels; and

determining, through a classifier model, whether a pulmonary function of the respective lung lobe is normal or abnormal based on the descriptors,

wherein the radiodensity values are represented in Hounsfield units, and wherein determining a plurality of regions of interest (ROIs) within each chest image further comprises removing protein and adipose from each chest image.

2 . The method of claim 1 , wherein determining the pulmonary function of the respective lung lobe further comprises:

determining that the pulmonary function of the respective lung lobe is normal in response to an average of the corresponding descriptors being greater than a threshold; and

determining that the pulmonary function of the respective lung lobe is abnormal in response to the average of the corresponding descriptors being smaller than the threshold.

3 . The method of claim 1 , wherein determining the pulmonary function of the respective lung lobe further comprises:

determining whether the pulmonary function of the respective lung lobe is normal or abnormal through applying the descriptors to the classifier model.

4 . The method of claim 3 , wherein the classifier model includes at least one of a support vector machine, a decision tree, a neural network, a random forest, or a regression model.

5 . The method of claim 3 , wherein each descriptor is further associated with at least one of: a serial number of ROI, a serial number of chest image, or coordinate values of the one or more pixels.

6 . The method of claim 3 , wherein the T ij descriptors for the j-th ROI of the i-th chest image are determined based on one of: the scale-invariant feature transform (SIFT) algorithm, the dense trajectory (DT) algorithm, or the improved dense trajectory (iDT) algorithm.

7 . The method of claim 3 , further comprising:

determining a feature vector for the j-th ROI of the i-th chest image based on the T ij descriptors; and

determining whether the pulmonary function is normal or abnormal through applying the feature vectors to the classifier model.

8 . The method of claim 1 , further comprising: determining whether a value of FEV1/FVC is greater or less than 70% through applying the descriptors to the classifier model.

9 . The method of claim 1 , wherein the image processing model includes a U-NET model.

10 . A device for detecting pulmonary function based on low-dose computed tomography (CT) chest images, comprising:

a processor; and

a memory coupled with the processor,

wherein the processor executes computer-readable instructions stored in the memory to perform operations, and the operations comprise:

receiving a plurality of chest images, the plurality of chest image generated by a low-dose CT method;

determining, by the processor, a plurality of regions of interest (ROIs) within each chest image through applying the plurality of chest images to an image processing model, the plurality of ROIs corresponding to five lung lobes;

determining, by the processor, T ij descriptors for the j-th ROI of the i-th chest image, each descriptor associated with a radiodensity value of one or more pixels; and

determining, by the processor, whether a pulmonary function of the respective lung lobe is normal or abnormal based on the descriptors through a classifier model,

wherein the radiodensity values are represented in Hounsfield units, and wherein determining a plurality of regions of interest (ROIs) within each chest image further comprises removing protein and adipose from each chest image.

11 . The device of claim 10 , wherein determining the pulmonary function of the respective lung lobe further comprises:

determining that the pulmonary function of the respective lung lobe is normal in response to an average of the corresponding descriptors being greater than a threshold; and

determining that the pulmonary function of the respective lung lobe is abnormal in response to the average of the corresponding descriptors being smaller than the threshold.

12 . The device of claim 10 , wherein determining the pulmonary function of the respective lung lobe further comprises:

determining whether the pulmonary function of the respective lung lobe is normal or abnormal through applying the descriptors to the classifier model.

13 . The device of claim 12 , wherein the classifier model includes at least one of a support vector machine, a decision tree, a neural network, a random forest, or a regression model.

14 . The device of claim 12 , wherein each descriptor is further associated with at least one of: a serial number of ROI, a serial number of chest image, or coordinate values of the one or more pixels.

15 . The device of claim 12 , wherein the T ij descriptors for the j-th ROI of the i-th chest image are determined based on one of: the scale-invariant feature transform (SIFT) algorithm, the dense trajectory (DT) algorithm, or the improved dense trajectory (iDT) algorithm.

16 . The device of claim 12 , wherein the operations further comprise:

determining a feature vector for the j-th ROI of the i-th chest image based on the T ij descriptors; and

determining whether the pulmonary function is normal or abnormal through applying the feature vectors to the classifier model.

17 . The device of claim 10 , wherein the classifier model is configured to determine whether a value of FEV1/FVC is greater or less than 70% based on the descriptors.

18 . The device of claim 10 , wherein the image processing model include a U-NET model.