Systems and method receiving image patches, and determining tissue sample biomarker status in digitized pathology samples using trained machine learning model

1 . A method comprising:

receiving a plurality of image patches corresponding to an image of a pathology slide having a hematoxylin and eosin-stained (“H&E”) stained sample of tissue, each image patch representing a different portion of the image;

for each image patch, determining, using a first trained machine learning (“ML”) model, a patch biomarker status;

for each patch biomarker status, comparing the respective patch biomarker status to a threshold,

responsive to determining that the respective patch biomarker status for biomarker positivity satisfies the threshold, determining the patch biomarker status as comprising biomarker positive;

responsive to determining that the respective patch biomarker status for biomarker negativity satisfies the threshold, determining the patch biomarker status as comprising biomarker negative; and

determining, using a second trained ML model, a tissue sample biomarker status for the sample of tissue based on the patch biomarker statuses of the image patches.

2 . The method of claim 1 , wherein:

the patch biomarker status comprises one or more of estrogen receptor (“ER”) positive, ER negative, progesterone receptor (“PR”) positive, PR negative, human epidermal growth factor receptor 2 (“HER2”) positive, HER2 negative, or not invasive carcinoma, and

the tissue sample biomarker status comprising one or more of ER positive, ER negative, PR positive, PR negative, HER2 positive, HER2 negative, or not invasive carcinoma.

3 . The method of claim 1 , wherein the patch biomarker status is a first patch biomarker status and the tissue sample biomarker status is a first tissue sample biomarker status, and further comprising:

for each image patch, determining, using a third trained ML model, a second patch biomarker status;

determining, using a fourth trained ML model, a second tissue sample biomarker status for the sample of tissue based on the second patch biomarker statuses of the image patches;

for each image patch, determining, using a fifth trained ML model, a third patch biomarker status; and

determining, using a sixth trained ML model, a third tissue sample biomarker status for the sample of tissue based on the third patch biomarker statuses of the image patches.

4 . The method of claim 3 , wherein:

the first patch biomarker status comprises one of estrogen receptor (“ER”) positive, ER negative, or not invasive carcinoma,

the first tissue sample biomarker status comprises an ER positive status, an ER negative status, and a not invasive carcinoma status,

the second patch biomarker status comprises a progesterone receptor (“PR”) positive status, a PR negative status, and a not invasive carcinoma status,

the second tissue sample biomarker status comprises one or more of a PR positive status, a PR negative status, or a not invasive carcinoma status,

the third patch biomarker status comprises a human epidermal growth factor receptor 2 (“HER2”) positive status, a HER2 negative status, and a not invasive carcinoma status,

the third tissue sample biomarker status comprises one or more of a HER2 positive status, a HER2 negative status, or a not invasive carcinoma status.

5 . The method of claim 1 , further comprising generating one or more histograms based on the patch biomarker statuses of the image patches, and wherein determining, using the second trained ML model, the tissue sample biomarker status for the sample of tissue is based on the one or more histograms.

6 . The method of claim 1 , wherein each patch biomarker status comprises a probability between 0% and 100%, inclusive, corresponding to a biomarker.

7 . The method of claim 1 , wherein the tissue sample biomarker status for the sample of tissue comprises a probability between 0% and 100%, inclusive, corresponding to a biomarker.

8 . The method of claim 1 ,

wherein the threshold is t;

responsive to determining a respective patch biomarker status for biomarker positivity is greater than or equal to t, determining the patch biomarker status as comprising biomarker positive;

responsive to determining a respective patch biomarker status for biomarker negativity is greater than or equal to t, determining the patch biomarker status as comprising biomarker negative; and

responsive to determining that a respective patch biomarker status for biomarker positivity is less than t and that the respective patch biomarker status for biomarker negativity is less than t, determining the patch biomarker status as comprising not invasive carcinoma.

9 . A system comprising:

a non-transitory computer-readable medium; and

one or more processors communicatively coupled to the non-transitory computer-readable medium, the one or more processors configured to execute processor executable instructions stored in the non-transitory computer-readable medium configured to cause the one or more processors to:

receive a plurality of image patches corresponding to an image of a pathology slide having a hematoxylin and eosin-stained (“H&E”) stained sample of tissue, each image patch representing a different portion of the image;

for each image patch, determine, using a first trained machine learning (“ML”) model, a patch biomarker status;

for each patch biomarker status, compare the respective patch biomarker status to a threshold,

responsive to a determination that the respective patch biomarker status for biomarker positivity satisfies the threshold, determine the patch biomarker status as comprising biomarker positive;

responsive to a determination that the respective patch biomarker status for biomarker negativity satisfies the threshold, determine the patch biomarker status as comprising biomarker negative; and

determine, using a second trained ML model, a tissue sample biomarker status for the sample of tissue based on the patch biomarker statuses of the image patches.

10 . The system of claim 9 , wherein:

the patch biomarker status comprises one of estrogen receptor (“ER”) positive, ER negative, progesterone receptor (“PR”) positive, PR negative, human epidermal growth factor receptor 2 (“HER2”) positive, HER2 negative, or not invasive carcinoma, and

the tissue sample biomarker status comprises one of ER positive, ER negative, PR positive, PR negative, HER2 positive, HER2 negative, or not invasive carcinoma.

11 . The system of claim 9 , wherein the patch biomarker status is a first patch biomarker status and the tissue sample biomarker status is a first tissue sample biomarker status, and wherein the one or more processors are configured to execute further processor executable instructions stored in the non-transitory computer-readable medium configured to cause the one or more processors to:

for each image patch, determine, using a third trained ML model, a second patch biomarker status;

determine, using a fourth trained ML model, a second tissue sample biomarker status for the sample of tissue based on the second patch biomarker statuses of the image patches;

for each image patch, determine, using a fifth trained ML model, a third patch biomarker status; and

determine, using a sixth trained ML model, a third tissue sample biomarker status for the sample of tissue based on the third patch biomarker statuses of the image patches.

12 . The system of claim 11 , wherein:

the first patch biomarker status comprises one of estrogen receptor (“ER”) positive, ER negative, or not invasive carcinoma,

the first tissue sample biomarker status comprises one of ER positive, ER negative, or not invasive carcinoma,

the second patch biomarker status comprises one of progesterone receptor (“PR”) positive, PR negative, or not invasive carcinoma,

the second tissue sample biomarker status comprises one of PR positive, PR negative, or not invasive carcinoma,

the third patch biomarker status comprises one of human epidermal growth factor receptor 2 (“HER2”) positive, HER2 negative, or not invasive carcinoma,

the third tissue sample biomarker status comprises one of HER2 positive, HER2 negative, or not invasive carcinoma.

13 . The system of claim 9 , wherein the one or more processors are configured to execute further processor executable instructions stored in the non-transitory computer-readable medium configured to cause the one or more processors to generate a histogram based on the patch biomarker statuses of the image patches, and determine, using the second trained ML model, the tissue sample biomarker status for the sample of tissue based on the histogram.

14 . The system of claim 9 , wherein each patch biomarker status comprises a probability between 0% and 100%, inclusive.

15 . The system of claim 9 , wherein the tissue sample biomarker status for the sample of tissue comprises a probability between 0% and 100%, inclusive.

16 . The system of claim 9 , wherein

the threshold is t; and wherein the one or more processors are configured to execute further processor executable instructions stored in the non-transitory computer-readable medium configured to cause the one or more processors to:

responsive to a determination that a respective patch biomarker status for biomarker positivity is greater than or equal to t, determine the patch biomarker status as comprising biomarker positive;

responsive to a determination a respective patch biomarker status for biomarker negativity is greater than or equal to t, determine the patch biomarker status as comprising biomarker negative; and

responsive to a determination that a respective patch biomarker status for biomarker positivity is less than t and that the respective patch biomarker status for biomarker negativity is less than t, determine the patch biomarker status as comprising not invasive carcinoma.

17 . A non-transitory computer-readable medium comprising processor-executable instructions configured to cause a processor to:

receive a plurality of image patches corresponding to an image of a pathology slide having a hematoxylin and eosin-stained (“H&E”) stained sample of tissue, each image patch representing a different portion of the image;

for each image patch, determine, using a first trained machine learning (“ML”) model, a patch biomarker status;

for each patch biomarker status, compare the respective patch biomarker status to a threshold,

responsive to a determination that the respective patch biomarker status for biomarker positivity satisfies the threshold, determine the patch biomarker status as comprising biomarker positive;

responsive to a determination that the respective patch biomarker status for biomarker negativity satisfies the threshold, determine the patch biomarker status as comprising biomarker negative; and

determine, using a second trained ML model, a tissue sample biomarker status for the sample of tissue based on the patch biomarker statuses of the image patches.

18 . The non-transitory computer-readable medium of claim 17 , wherein:

the patch biomarker status comprises one of estrogen receptor (“ER”) positive, ER negative, progesterone receptor (“PR”) positive, PR negative, human epidermal growth factor receptor 2 (“HER2”) positive, HER2 negative, or not invasive carcinoma, and

the tissue sample biomarker status comprises one of ER positive, ER negative, PR positive, PR negative, HER2 positive, HER2 negative, or not invasive carcinoma.

19 . The non-transitory computer-readable medium of claim 17 , wherein the patch biomarker status is a first patch biomarker status and the tissue sample biomarker status is a first tissue sample biomarker status, and further comprising processor-executable instructions configured to cause the processor to:

for each image patch, determine, using a third trained ML model, a second patch biomarker status;

determine, using a fourth trained ML model, a second tissue sample biomarker status for the sample of tissue based on the second patch biomarker statuses of the image patches;

for each image patch, determine, using a fifth trained ML model, a third patch biomarker status; and

determine, using a sixth trained ML model, a third tissue sample biomarker status for the sample of tissue based on the third patch biomarker statuses of the image patches.

20 . The non-transitory computer-readable medium of claim 19 , wherein:

the first patch biomarker status comprises one of estrogen receptor (“ER”) positive, ER negative, or not invasive carcinoma,

the first tissue sample biomarker status comprises one of ER positive, ER negative, or not invasive carcinoma,

the second patch biomarker status comprises one of progesterone receptor (“PR”) positive, PR negative, or not invasive carcinoma,

the second tissue sample biomarker status comprises one of PR positive, PR negative, or not invasive carcinoma,

the third patch biomarker status comprises one of human epidermal growth factor receptor 2 (“HER2”) positive, HER2 negative, or not invasive carcinoma,

the third tissue sample biomarker status comprises one of HER2 positive, HER2 negative, or not invasive carcinoma.