Case Western Reserve University

Rutgers, the State University of New Jersey

The Cleveland Clinic Foundation

Andrew Janowczyk

Anant Madabhushi

Nathaniel Braman

Patrick Leo

Kavya Ravichandran

Kaustav Bera

Michael D Feldman

Sanjay Gupta

Yijiang Chen

Eliot G Peyster

Kenneth B Margulies

Sharat Chandran

Laura Barisoni

John Sedor

Jeffrey John Nirschl

Catherine Jayapandian

The present disclosure relates to a non-transitory computer-readable medium storing computer-executable instructions that, when executed, cause a processor to perform operations. The operations include extracting one or more image characterization metrics from respective ones of a plurality of digitized images within an imaging data set. The plurality of digitized images have batch effects. The operations further include identifying a plurality of batch effect groups of the digitized images using the one or more image characterization metrics, and dividing the plurality of batch effect groups between a training set and/or a validation set. The training set and/or the validation set include some of the plurality of digitized images associated with respective ones of the plurality of batch effect groups.

Batch effect mitigation in digitized images

Embodiments include accessing a set of digital pathology (DP) images having an imaging parameter; applying a low-computational cost histology quality control (HistoQC) pipeline to the DP images, where the low-computational cost HistoQC pipeline computes a first set of image metrics associated with a DP image, and assigns the DP image to a first or a second, different cohort based on the imaging parameter and the first set of image metrics; applying a first, higher-computational-cost HistoQC pipeline to a member of the first cohort; applying a second, different higher-computation-cost HistoQC pipeline to a member of the second cohort; where the first or second, higher-computational-cost HistoQC pipeline determines an artifact-free region of the member of the first or second cohort, respectively, and classifies the member of the first or second cohort, respectively, as suitable or unsuitable for downstream computation or diagnostic analysis based, at least in part, on the artifact free region.

Quality control for digital pathology slides

Embodiments predict response to neoadjuvant chemotherapy (NAC) in breast cancer (BCa) from pre-treatment dynamic contrast enhanced magnetic resonance imaging (DCE-MRI). Embodiments compute, using a machine learning (ML) classifier, a first probability of response based on a set of radiomic features extracted from a tumoral region represented in a pre-treatment DCE-MRI image of a region of tissue (ROT) demonstrating BCa; extract patches from the tumoral region; provide the patches to a convolutional neural network (CNN); receive, from the CNN, a pixel-level localized patch probability of response; compute a second probability of response based on the pixel-level localized patch probability; compute a combined ML probability from the first and second probabilities; compute a final probability of response based on the combined ML probability and clinical information associated with the ROT; classify the ROT as a responder or non-responder based on the final probability of response; and display the classification.

Predicting neo-adjuvant chemotherapy response from pre-treatment breast magnetic resonance imaging using artificial intelligence and HER2 status

Embodiments discussed herein facilitate segmentation of histological primitives from stained histology of renal biopsies via deep learning and/or training deep learning model(s) to perform such segmentation. One example embodiment is configured to access a first histological image of a renal biopsy comprising a first type of histological primitives, wherein the first histological image is stained with a first type of stain; provide the first histological image to a first deep learning model trained based on the first type of histological primitive and the first type of stain; and receive a first output image from the first deep learning model, wherein the first type of histological primitives is segmented in the first output image.

Deep learning-based multi-site, multi-primitive segmentation for nephropathology using renal biopsy whole slide images

Embodiments facilitate generating a biochemical recurrence (BCR) prognosis by accessing a digitized image of a region of tissue demonstrating prostate cancer (CaP) pathology associated with a patient; generating a set of segmented gland lumen by segmenting a plurality of gland lumen represented in the region of tissue using a deep learning segmentation model; generating a set of post-processed segmented gland lumen; extracting a set of quantitative histomorphometry (QH) features from the digitized image based, at least in part, on the set of post-processed segmented gland lumen; generating a feature vector based on the set of QH features; computing a histotyping risk score based on a weighted sum of the feature vector; generating a classification of the patient as BCR high-risk or BCR low-risk based on the histotyping risk score and a risk score threshold; generating a BCR prognosis based on the classification; and displaying the BCR prognosis.

Prognosis of prostate cancer with computerized histomorphometric features of tumor morphology from routine hematoxylin and eosin slides

Embodiments access a pre-neoadjuvant chemotherapy (NAC) radiological image of a region of tissue demonstrating breast cancer (BCa), the region of tissue including a tumoral region, the image having a plurality of pixels; extract a set of patches from the tumoral region; provide the set of patches to a convolutional neural network (CNN) configured to discriminate tissue that will experience pathological complete response (pCR) post-NAC from tissue that will not; receive, from the CNN, a pixel-level localized patch probability of pCR; compute a distribution of predictions across analyzed patches based on the pixel-level localized patch probability; classify the region of tissue as a responder or non-responder based on the distribution of predictions, and display the classification. Embodiments may further generate a probability mask based on the pixel-level localized patch probability; and generate a heatmap of likelihood of response to NAC based on the probability mask and the pre-NAC radiological image.

Predicting pathological complete response to neoadjuvant chemotherapy from baseline breast dynamic contrast enhanced magnetic resonance imaging (DCE-MRI)

Embodiments include controlling a processor to perform operations for predicting biochemical recurrence (BCR) in prostate cancer (PCa), including accessing a first digitized pathology slide having a first stain channel of a region of tissue demonstrating PCa; accessing a second digitized pathology slide having a second, different stain channel of the region of tissue; extracting morphology features from the first stain channel; extracting stain intensity features from the second stain channel, where a stain intensity feature quantifies an amount of a molecular biomarker present in a cellular nucleus; controlling a first machine learning classifier to generate a first probability of BCR based on the morphology features; controlling a second machine learning classifier to generate a second, different probability of BCR based on the stain intensity features; computing an aggregate probability of BCR based on the first probability and the second probability; and displaying the aggregate probability.

Predicting prostate cancer biochemical recurrence using combined nuclear NF-KB/P65 localization and gland morphology

Methods, apparatus, and other embodiments predict heart failure from WSIs of cardiac histopathology using a deep learning convolutional neural network (CNN). One example apparatus includes a pre-processing circuit configured to generate a pre-processed WSI by downsampling a digital WSI; an image acquisition circuit configured to randomly select a set of non-overlapping ROIs from the pre-processed WSI, and configured to provide the set of non-overlapping ROIs to a deep learning circuit; a deep learning circuit configured to generate an image-level probability that a member of the set of non-overlapping ROIs is a failure/abnormal pathology ROI using a CNN; and a classification circuit configured to generate a patient-level probability that the patient from which the region of tissue represented in the WSI was acquired is experiencing failure or non-failure based, at least in part, on the image-level probability.

Histomorphometric classifier to predict cardiac failure from whole-slide hematoxylin and eosin stained images

A method and apparatus for obtaining segmented images of the stained regions may comprise quantifying the extent of the presence of staining of a biomarker in an original image of a sample, which may comprise selecting a domain swatch of data based upon a user specified domain knowledge; clustering the data within the original image by conducting a frequency weighted mean shift of the data within the original image to convergence, forming a hierarchical plurality of layers each having a different data resolution to form a hierarchical data pyramid; segmenting the plurality of mean shifted data images to determine in each mean shifted data image within the hierarchical data pyramid data not excluded as outside of the swatch; mapping the data not excluded as outside the swatch spatially back to the original image to create a final image; and, storing the final image on a storage medium for further analysis.

Growing community of inventors

East Meadow, NY, United States of America

Andrew Janowczyk