Systems and methods for polyp size estimation

1 . A computer-implemented method for estimating a size of a polyp without human intervention using semantic segmentation by deep learning convolutional neural networks for polyp detection and localization followed by region property analysis to determine extreme peripheral points of the estimated polyp followed by size estimation, the computer-implemented method comprising:

accessing a number k of light images of at least a portion of a gastrointestinal tract (GIT) captured during a capsule endoscopy procedure using light, wherein k>1, wherein each light image of the unstructured light images is associated with a respective estimated size of a suspected polyp determined based on the respective light image, the estimated sizes of the suspected polyp comprising a first size value size 1 and a second size value size 2 , wherein the light images are determined to include the same suspected polyp;

determining a weight of each light image of the light images indicative of a probability of the respective light image including a polyp, the weights comprising a first weight value weight 1 and a second weight value weight 2 ; and

estimating a weighted polyp size for the suspected polyp based on the determined weights of the light images and based on the estimated sizes of the suspected polyp according to the light images,

wherein the estimating the weighted polyp size comprises computing:

(weight 1 ·size 1 +weight 2 ·size 2 ).

2 . The computer-implemented method of claim 1 , further comprising identifying the accessed light images from a plurality of images of the GIT based on a machine learning system.

3 . The computer-implemented method of claim 1 , further comprising estimating the size of the suspected polyp in each light image of the light images that includes the suspected polyp.

4 . The computer-implemented method of claim 1 , wherein the weight of each light image of the light images is determined further based on a distance of the light image from a predefined light image.

5 . The computer-implemented method of claim 1 , wherein the weight of each light image of the light images is determined further based on a cleansing score of the light image.

6 . The computer-implemented method of claim 1 , wherein the weight of each light image of the light images is determined further based on: a distance of the light image from a predefined light image, and a cleansing score of the light image.

7 . The computer-implemented method of claim 1 ,

wherein the estimated sizes of the suspected polyp comprise a third size value size 3 ,

wherein the weights comprise a third weight value weight 3 , and

wherein the estimating the weighted polyp size further comprises computing:

(weight 3 ·size 3 ),

such that the estimating the weighted polyp size comprises computing:

(weight 1 ·size 1 +weight 2 ·size 2 +weight 3 ·size 3 ).

8 . An apparatus for estimating a size of a polyp without human intervention using semantic segmentation by deep learning convolutional neural networks for polyp detection and localization followed by region property analysis to determine extreme peripheral points of the estimated polyp followed by size estimation, the apparatus comprising:

at least one processor; and

at least one memory having stored thereon instructions which, when executed by the at least one processor, cause the apparatus at least to perform:

accessing a number k of light images of at least a portion of a gastrointestinal tract (GIT) captured during a capsule endoscopy procedure using light, wherein k>1, wherein each light image of the light images is associated with a respective estimated size of a suspected polyp determined based on the respective light image, the estimated sizes of the suspected polyp comprising a first size value size 1 and a second size value size 2 , wherein the light images are determined to include the same suspected polyp;

determining a weight of each light image of the light images indicative of a probability of the respective light image including a polyp, the weights comprising a first weight value weight 1 and a second weight value weight 2 ; and

estimating a weighted polyp size for the suspected polyp based on the determined weights of the light images and based on the estimated sizes of the suspected polyp according to the light images,

wherein the estimating the weighted polyp size comprises computing:

(weight 1 ·size 1 +weight 2 ·size 2 ).

9 . A computer-readable medium having stored thereon instructions for estimating a size of a polyp without human intervention using semantic segmentation by deep learning convolutional neural networks for polyp detection and localization followed by region property analysis to determine extreme peripheral points of the estimated polyp followed by size estimation, the instructions, when executed by at least one processor of an apparatus, cause the apparatus at least to perform:

accessing a number k of light images of at least a portion of a gastrointestinal tract (GIT) captured during a capsule endoscopy procedure using light, wherein k>1, wherein each light image of the light images is associated with a respective estimated size of a suspected polyp determined based on the respective light image, the estimated sizes of the suspected polyp comprising a first size value size 1 and a second size value size 2 , wherein the light images are determined to include the same suspected polyp;

determining a weight of each light image of the light images indicative of a probability of the respective light image including a polyp, the weights comprising a first weight value weight 1 and a second weight value weight 2 ; and

estimating a weighted polyp size for the suspected polyp based on the determined weights of the light images and based on the estimated sizes of the suspected polyp according to the light images,

wherein the estimating the weighted polyp size comprises computing:

(weight 1 ·size 1 +weight 2 ·size 2 ).