System for diagnosing Barrett's esophagus and related methods

1 . A system for diagnosing Barrett's esophagus (BE) in a patient, the system comprising:

a processor and memory coupled thereto, the processor configured to

receive a whole slide image (WSI) from an esophagus sample from the patient,

segment the WSI into a plurality of image segments,

extract a plurality of features from each image segment using a convolutional neural network (CNN),

assign each image segment an attention score based upon the plurality of features,

filter the plurality of image segments based upon the attention score of each image segment, the filtering comprising

clustering the plurality of image segments based upon respective attention scores and upper and lower thresholds, and

removing a portion of the image segments with attention scores between the upper and lower thresholds,

generate at least one BE metric value for the patient based upon the filtered image segments, the at least one BE metric value indicating a probability that tissue with a propensity to progress to BE exists within the WSI.

2 . The system of claim 1 wherein the assigning comprises ingesting the plurality of features for each image segment into an attention network.

3 . The system of claim 1 wherein the assigning comprises generating a plurality of branches, each branch associated with a respective class.

4 . The system of claim 3 wherein at least one branch is associated with a slide-level ground truth label class.

5 . The system of claim 1 wherein the filtering comprises

ingesting the plurality of features for each image segment into a cluster network.

6 . The system of claim 1 wherein the filtering comprises removing image segments having less than a threshold amount of tissue.

7 . The system of claim 1 wherein the at least one BE metric value comprises a plurality thereof, the plurality of BE metric values comprising a first BE metric indicating a probability that tissue with a propensity to progress to BE exists within the WSI.

8 . The system of claim 1 wherein the processor is configured to directly ingest the plurality of image segments into the CNN.

9 . The system of claim 1 wherein the extracting comprises placing the plurality of features from each image segment in a data space less than a data space of the WSI.

10 . A system for diagnosing Barrett's esophagus (BE) in a patient, the system comprising:

a processor and memory coupled thereto, the processor configured to

receive a whole slide image (WSI) from an esophagus sample from the patient,

segment the WSI into a plurality of image segments,

directly ingest the plurality of image segments into a convolutional neural network (CNN),

extract a plurality of features from each image segment using the CNN,

assign each image segment an attention score based upon the plurality of features,

filter the plurality of image segments based upon the attention score of each image segment, the filtering comprising

ingesting the plurality of features for each image segment into a cluster network,

clustering the plurality of image segments with attention scores based upon respective attention scores and upper and lower thresholds, and

removing a portion of the image segments with attention scores between the first and second thresholds, and

generate at least one BE metric value for the patient based upon the filtered image segments, the at least one BE metric value indicating a probability that tissue with a propensity to progress to BE exists within the WSI.

11 . The system of claim 10 wherein the assigning comprises ingesting the plurality of features for each image segment into an attention network.

12 . The system of claim 10 wherein the assigning comprises generating a plurality of branches, each branch associated with a respective class.

13 . The system of claim 12 wherein at least one branch is associated with a slide-level ground truth label class.

14 . The system of claim 10 wherein the filtering comprises removing image segments having less than a threshold amount of tissue.

15 . The system of claim 10 wherein the at least one BE metric value comprises a plurality thereof, the plurality of BE metric values comprising a first BE metric indicating a probability that tissue with a propensity to progress to BE exists within the WSI.

16 . The system of claim 10 wherein the extracting comprises placing the plurality of features from each image segment in a data space less than a data space of the WSI.

17 . A method for diagnosing Barrett's esophagus (BE) in a patient, the method comprising:

receiving a whole slide image (WSI) from an esophagus sample from the patient;

segmenting the WSI into a plurality of image segments;

extracting a plurality of features from each image segment using a convolutional neural network (CNN);

assigning each image segment an attention score based upon the plurality of features;

filtering the plurality of image segments based upon the attention score of each image segment, the filtering comprising

clustering the plurality of image segments based upon respective attention scores and upper and lower thresholds, and

removing a portion of the image segments with attention scores between the upper and lower thresholds; and

generating at least one BE metric value for the patient based upon the filtered image segments, the at least one BE metric value indicating a probability that tissue with a propensity to progress to BE exists within the WSI.

18 . The method of claim 17 wherein the assigning comprises ingesting the plurality of features for each image segment into an attention network.

19 . The method of claim 17 wherein the assigning comprises generating a plurality of branches, each branch associated with a respective class.

20 . The method of claim 19 wherein at least one branch is associated with a slide-level ground truth label class.