Case Western Reserve University

The United States Government As Represented by the Department of Veterans Affairs

Tempus Ai, Inc.

Nathaniel Braman

Anant Madabhushi

Mirabela Rusu

Mahdi Orooji

Robert Gilkeson

Philip Linden

Andrew Janowczyk

Prateek Prasanna

Mehdi Alilou

Kavya Ravichandran

Jagadish Venkataraman

Lyndsay Harris

Vinay Varadan

Yijiang Chen

Tristan Maidment

Emery T Goossens

Salendra Singh

Jeffrey Eben

The present disclosure relates to a method. The method may be performed by accessing data derived from one or more routine clinical medical imaging scans including a lesion in which the lesion and associated vasculature are segmented in a three-dimensional segmentation. At least two features are extracted from the three-dimensional segmentation of the associated vasculature. The at least two features include at least one feature indicative of a morphology of the associated vasculature or a portion thereof, and at least one feature indicative of a function of the associated vasculature or a portion thereof. The at least two features, and/or one or more statistics of the at least two features, are provided to a machine learning model trained to make a prediction concerning the lesion. The prediction concerning the lesion is received from the machine learning model.

Tumor characterization and outcome prediction through quantitative measurements of tumor-associated vasculature

A system and method are provided for identifying a multimodal biomarker of a prognostic prediction, using a deep learning framework trained to analyze different modality data, including radiomic image data, pathology image data, and molecular image data to obtain unimodal embedding predictions from those modality data and generate multimodal embedding predictions, through application of a loss minimization and attention-based fusion processes.

Systems and methods for deep orthogonal fusion for multimodal prognostic biomarker discovery

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 determination of a response to treatment and/or a prognosis for a tumor based at least in part on features of tumor-associated vasculature (TAV). One example embodiment is a method, comprising: accessing a medical imaging scan of a tumor, wherein the tumor is segmented on the medical imaging scan; segmenting tumor-associated vasculature (TAV) associated with the tumor based on the medical imaging scan; extracting one or more features from the TAV; providing the one or more features extracted from the TAV to a trained machine learning model; and receiving, from the machine learning model, one of a predicted response to a treatment for the tumor or a prognosis for the tumor.

Embodiments discussed herein facilitate determination of whether lesions are benign or malignant. One example embodiment is a method, comprising: accessing medical imaging scan(s) that are each associated with distinct angle(s) and each comprise a segmented region of interest (ROI) of that medical imaging scan comprising a lesion associated with a first region and a second region; providing the first region(s) of the medical imaging scan(s) to trained first deep learning (DL) model(s) of an ensemble and the second region(s) of the medical imaging scan(s) to trained second DL model(s) of the ensemble; and receiving, from the ensemble of DL models, an indication of whether the lesion is a benign architectural distortion (AD) or a malignant AD.

Specialized computer-aided diagnosis and disease characterization with a multi-focal ensemble of convolutional neural networks

Embodiments discussed herein facilitate training and/or employing a combined model employing machine learning and deep learning outputs to generate prognoses for treatment of tumors. One example embodiment can extract radiomic features from a tumor and a peri-tumoral region; provide the intra-tumoral and peri-tumoral features to two separate machine learning models; provide the segmented tumor and peri-tumoral region to two separate deep learning models; receive predicted prognoses from each of the machine learning models and each of the deep learning models; provide the predicted prognoses to a combined machine learning model; and receive a combined predicted prognosis for the tumor from the combined machine learning model.

Disease characterization and response estimation through spatially-invoked radiomics and deep learning fusion

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 access a radiological image of tissue having a tumoral volume and a peritumoral volume; define a vasculature associated with the tumoral volume; generate a Cartesian two-dimensional (2D) vessel network representation; compute a first set of localized Hough transforms based on the Cartesian 2D vessel network representation; generate a first aggregated set of peak orientations based on the first set of Hough transforms; generate a spherical 2D vessel network representation; compute a second set of localized Hough transforms based on the spherical 2D vessel network representation; generate a second aggregated set of peak orientations based on the second set of Hough transforms; generate a vascular network organization descriptor based on the aggregated peak orientations; compute a probability that the tissue is a member of a positive class based on the vascular network organization descriptor; classify the ROI based on the probability; and display the classification.

Vascular network organization via Hough transform (VaNgOGH): a radiomic biomarker for diagnosis and treatment response

Embodiments associated with classifying a region of tissue using features extracted from nodules and surrounding structures. One example apparatus includes a feature extraction circuit configured to automatically extract a first set of quantitative features from a nodule represented in at least one CT image, and automatically extract a second set of quantitative features from the lung parenchyma region immediately surrounding the nodule represented in the at least one CT image; a feature selection circuit configured to select an optimally predictive feature set from the first set of quantitative features and the second set of quantitative features; and a training circuit configured to train a classifier using the optimally predictive feature set to assign malignancy risk to a lung nodule represented in a CT image of a region of tissue demonstrating lung nodules. A prognosis or treatment plan may be provided based on the malignancy risk.

Characterizing lung nodule risk with quantitative nodule and perinodular radiomics

Methods, apparatus, and other embodiments associated with classifying a region of tissue using textural analysis are described. One example apparatus includes an image acquisition logic that acquires an image of a region of tissue demonstrating cancerous pathology, a delineation logic that distinguishes nodule tissue within the image from the background of the image, a perinodular zone logic that defines a perinodular zone based on the nodule, a feature extraction logic that extracts a set of features from the image, a probability logic that computes a probability that the nodule is benign or that the nodule will respond to a treatment, and a classification logic that classifies the nodule tissue based, at least in part, on the set of features or the probability. A prognosis or treatment plan may be provided based on the classification of the image.

Decision support for disease characterization and treatment response with disease and peri-disease radiomics

Methods, apparatus, and other embodiments associated with classifying a region of tissue using quantified vessel tortuosity are described. One example apparatus includes an image acquisition logic that acquires an image of a region of tissue demonstrating cancerous pathology, a delineation logic that distinguishes nodule tissue within the image from the background of the image, a perinodular zone logic that defines a perinodular zone based on the nodule, a feature extraction logic that extracts a set of features from the image including a set of tortuosity features, a probability logic that computes a probability that the nodule is benign, and a classification logic that classifies the nodule tissue based, at least in part, on the set of features or the probability. A prognosis or treatment plan may be provided based on the classification of the image.

Characterizing disease and treatment response with quantitative vessel tortuosity radiomics

Methods, apparatus, and other embodiments distinguish disease phenotypes and mutational status using co-occurrence of local anisotropic gradient orientations (CoLIAGe) and Laws features. One example apparatus includes a set of circuits that acquires a radiologic image (e.g., MRI image) of a region of tissue demonstrating breast cancer, computes a gradient orientation for a pixel in the MRI image, computes a significant orientation for the pixel based on the gradient orientation, constructs a feature vector that captures a discretized entropy distribution for the image based on the significant orientation, extracts a set of texture features from the MRI image, and classifies the phenotype of the breast cancer based on the feature vector and the set of texture features. Embodiments of example apparatus may generate and display a heatmap of entropy values for the image. Example methods and apparatus may operate substantially in real-time, or may operate in two, three, or more dimensions.

Growing community of inventors

Cleveland, OH, United States of America

Nathaniel Braman