Systems and methods for machine learning (ML) model diagnostic assessments based on digital pathology data
Techniques for performing diagnostic assessments based on digital pathology data are disclosed. In one particular embodiment, the techniques may be realized as a method for performing a diagnostic assessment based on digital pathology data comprising obtaining first digital pathology data comprising intensity information, the first digital pathology data being associated with a plurality of regions of interest in a biological sample; applying first machine learning models to the first digital pathology data, the first machine learning models identifying first regions of interest among the plurality of regions of interest based on the intensity information; applying second machine learning models to the first digital pathology data, the second machine learning models identifying at least one pattern associated with at least one of the first regions of interest; generating a diagnostic assessment based on the first regions of interest and the at least one pattern.
1 . A method for performing a diagnostic assessment based on digital pathology data, comprising:
obtaining first digital pathology data comprising intensity information, the first digital pathology data being associated with a plurality of regions of interest in a biological sample;
applying one or more first machine learning models to the first digital pathology data, the one or more first machine learning models identifying one or more first regions of interest among the plurality of regions of interest based on the intensity information;
applying one or more second machine learning models to the first digital pathology data, the one or more second machine learning models identifying at least one pattern associated with at least one of the one or more first regions of interest;
generating a diagnostic assessment based on the one or more first regions of interest and the at least one pattern; and
detecting drift in diagnostic assessments performed by at least one pathologist in a clinical trial by comparing the generated diagnostic assessment with the pathologist diagnostic assessments, wherein detecting drift comprises detecting a change over time in at least one of nomenclature, grading of lesions, and scoring of a bio marker.
2 . The method of claim 1 , wherein the first digital pathology data comprises one or more whole slide images.
3 . The method of claim 2 , wherein the one or more whole slide images corresponds to a tumor biopsy sample stained using anti-HER2 immunohistochemistry.
4 . The method of claim 3 , wherein the diagnostic assessment comprises a HER2 score.
5 . The method of claim 1 , wherein the one or more first regions of interest comprises at least one of a tissue region of interest or a cell of interest.
6 . The method of claim 5 , wherein the at least one of the tissue region of interest or the cell of interest comprises one or more of a cancer epithelium, a cancer stroma, a ductal carcinoma in situ, a necrosis, a cell membrane, or an artifact.
7 . The method of claim 1 , wherein the least one pattern comprises a staining pattern of a cell membrane.
8 . The method of claim 7 , wherein the staining pattern is selected from a group consisting of: negative or unstained, partial positive, and complete positive.
9 . The method of claim 1 , wherein the diagnostic assessment comprises a precision slide-level score.
10 . The method of claim 1 , wherein the diagnostic assessment comprises an adjusted slide level score, the adjusted slide level score being generated using machine learning model predictions optimized for consensus between the adjusted slide level score and a slide level score provided by a pathologist.
11 . The method of claim 1 , further comprising applying one or more third machine learning models to the first digital pathology data, the one or more third machine learning models identifying an intensity associated with at least one of the one or more first regions of interest.
12 . The method of claim 11 , wherein the intensity corresponds to an intensity of staining of cell membranes, wherein the intensity is selected from a group consisting of: unstained, faintly stained, moderately stained, or completely stained.
13 . The method of claim 1 , further comprising extracting one or more histological features associated with the first digital pathology data.
14 . The method of claim 1 , further comprising calculating one or more cell-level features associated with the first digital pathology data.
15 . The method of claim 14 , wherein the one or more cell-level features are based on a number of cells corresponding to each American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) category identified in the first digital pathology data.
16 . The method of claim 1 , wherein the tumor biopsy sample is derived from a patient with breast cancer.
17 . A system for performing a diagnostic assessment based on digital pathology data comprising:
at least one computer processor, wherein the at least one computer processor is configured to:
obtain first digital pathology data comprising intensity information, the first digital pathology data being associated with a plurality of regions of interest in a biological sample;
apply one or more first machine learning models to the first digital pathology data, the one or more first machine learning models identifying one or more first regions of interest among the plurality of regions of interest based on the intensity information;
apply one or more second machine learning models to the first digital pathology data, the one or more second machine learning models identifying at least one pattern associated with at least one of the one or more first regions of interest;
generate a diagnostic assessment based on the one or more first regions of interest and the at least one pattern; and
detect drift in diagnostic assessments performed by at least one pathologist in a clinical trial by comparing the generated diagnostic assessment with the pathologist diagnostic assessments, wherein detecting drift comprises detecting a change over time in at least one of nomenclature, grading of lesions, and scoring of a bio marker.
18 . The system of claim 17 , wherein the first digital pathology data comprises one or more whole slide images, the one or more whole slide images corresponding to a tumor biopsy sample stained using anti-HER2 immunohistochemistry, and wherein the diagnostic assessment comprises a HER2 score.
19 . The system of claim 17 , wherein the one or more first regions of interest comprises at least one of a tissue region of interest or a cell of interest, the at least one of the tissue region of interest or the cell of interest comprising one or more of a cancer epithelium, a cancer stroma, a ductal carcinoma in situ, a necrosis, a cell membrane, or an artifact.
20 . The system of claim 17 , wherein the least one pattern comprises a staining pattern of a cell membrane, the staining pattern being selected from a group consisting of: negative or unstained, partial positive, and complete positive.
21 . The system of claim 17 , wherein the diagnostic assessment comprises:
a precision slide-level score; and
an adjusted slide level score, the adjusted slide level score being generated using machine learning model predictions optimized for consensus between the adjusted slide level score and a slide level score provided by a pathologist.
22 . The system of claim 17 , further comprising:
applying one or more third machine learning models to the first digital pathology data, the one or more third machine learning models identifying an intensity associated with at least one of the one or more first regions of interest, wherein the intensity corresponds to an intensity of staining of cell membranes, wherein the intensity is selected from a group consisting of: unstained, faintly stained, moderately stained, or completely stained;
extracting one or more histological features associated with the first digital pathology data; and
calculating one or more cell-level features associated with the first digital pathology data.
23 . An article of manufacture for performing a diagnostic assessment based on digital pathology data comprising:
a non-transitory processor readable medium; and
instructions stored on the medium;
wherein the instructions are configured to be readable from the medium by at least one computer processor and thereby cause the at least one computer processor to operate so as to:
obtain first digital pathology data comprising intensity information, the first digital pathology data being associated with a plurality of regions of interest in a biological sample;
apply one or more first machine learning models to the first digital pathology data, the one or more first machine learning models identifying one or more first regions of interest among the plurality of regions of interest based on the intensity information;
apply one or more second machine learning models to the first digital pathology data, the one or more second machine learning models identifying at least one pattern associated with at least one of the one or more first regions of interest;
generate a diagnostic assessment based on the one or more first regions of interest and the at least one pattern; and
detect drift in diagnostic assessments performed by at least one pathologist in a clinical trial by comparing the generated diagnostic assessment with the pathologist diagnostic assessments, wherein detecting drift comprises detecting a change over time in at least one of nomenclature, grading of lesions, and scoring of a bio marker.
24 . The article of manufacture of claim 23 , wherein the first digital pathology data comprises one or more whole slide images, the one or more whole slide images corresponding to a tumor biopsy sample stained using anti-HER2 immunohistochemistry, and wherein the diagnostic assessment comprises a HER2 score.
25 . The article of manufacture of claim 23 , wherein the one or more first regions of interest comprises at least one of a tissue region of interest or a cell of interest, the at least one of the tissue region of interest or the cell of interest comprising one or more of a cancer epithelium, a cancer stroma, a ductal carcinoma in situ, a necrosis, a cell membrane, or an artifact.
26 . The article of manufacture of claim 23 , wherein the least one pattern comprises a staining pattern of a cell membrane, the staining pattern being selected from a group consisting of: negative or unstained, partial positive, and complete positive.
27 . The article of manufacture of claim 23 , wherein the diagnostic assessment comprises:
a precision slide-level score; and
an adjusted slide level score, the adjusted slide level score being generated using machine learning model predictions optimized for consensus between the adjusted slide level score and a slide level score provided by a pathologist.
28 . The article of manufacture of claim 23 , further comprising:
applying one or more third machine learning models to the first digital pathology data, the one or more third machine learning models identifying an intensity associated with at least one of the one or more first regions of interest, wherein the intensity corresponds to an intensity of staining of cell membranes, wherein the intensity is selected from a group consisting of: unstained, faintly stained, moderately stained, or completely stained;
extracting one or more histological features associated with the first digital pathology data; and
calculating one or more cell-level features associated with the first digital pathology data.