University of South Florida

Stereology Resource Center, Inc.

H. Lee Moffitt Cancer Center and Research Institute, Inc.

Lawrence O Hall

Dmitry B Goldgof

Peter Randolph Mouton

Robert J Gillies

Baishali Chaudhury

Hady Ahmady Phoulady

Robert A Gatenby

Matthew B Schabath

Olya Stringfield

Rahul Paul

Yoganand Balagurunathan

Natarajan Raghunand

Kurt A Kramer

Daniel T Elozory

John Arrington

Om Pavithra Bonam

Palak Pankajbhai Dave

Systems and methods for automated stereology using deep learning are disclosed. The systems include an update in the form of a semi-automatic approach for ground truth preparation in 3D stacks of microscopy images (disector stacks) for generating more training data. The systems also present an exemplary disector-based MIMO framework where all the planes of a 3D disector stack are analyzed as opposed to a single focus-stacked image (EDF image) per stack. The MIMO approach avoids the costly computations of 3D deep learning-based methods by using the 3D context of cells in disector stacks; and prevents stereological bias in the previous EDF-based method due to counting profiles rather than cells and under-counting overlap-ping/occluded cells. Taken together, these improvements support the view that AI-based automatic deep learning methods can accelerate the efficiency of unbiased stereology cell counts without a loss of accuracy or precision as compared to conventional manual stereology.

Systems and methods for determining cell number count in automated stereology z-stack images

Systems and methods for automated stereology are provided. A method can include providing an imager for capturing a Z-stack of images of a three-dimensional (3D) object; constructing extended depth of field (EDF) images from the Z-stack of images; performing a segmentation method on the EDF images including estimating a Gaussian Mixture Model (GMM), performing morphological operations, performing watershed segmentation, constructing Voronoi diagrams and performing boundary smoothing; and determining one or more stereology parameters such as number of cells in a region.

Automated stereology for determining tissue characteristics

The present disclosure describes a multi-initialization ensemble-based defense strategy against an adversarial attack. In one embodiment, an exemplary method includes training a plurality of conventional neural networks (CNNs) with a training set of images, wherein the images include original images and images modified by an adversarial attack; after training of the plurality of conventional neural networks, providing an input image to the plurality of conventional neural networks, wherein the input image has been modified by an adversarial attack; receiving a probability output for the input image from each of the plurality of conventional neural networks; producing an ensemble probability output for the input image by combining the probability outputs from each of the plurality of conventional neural networks; and labeling the input image as belonging to one of the one or more categories based on the ensemble probability output.

Mitigating adversarial attacks on medical imaging understanding systems

Virtually every cancer patient is imaged with CT, PET or MRI. Importantly, such imaging reveals that tumors are complex and heterogeneous, often containing multiple habitats within them. Disclosed herein are methods for analyzing these images to infer cellular and molecular structure in each of these habitats. The methods can involve spatially superimposing two or more radiological images of the tumor sufficient to define regional habitat variations in two or more ecological dynamics in the tumor, and comparing the habitat variations to one or more controls to predict the severity of the tumor.

Radiologically identified tumor habitats

Systems and methods for applying an ensemble of segmentations to microscopy images of a tissue sample to determine if the tissue sample is representative of cancerous tissue. The ensemble of segmentations is applied to a plurality of greyscale or color microscopy images to generate a final image level segmentation and a final blob level segmentation. The final image level segmentation and final blob level segmentation are used to calculate a mean nuclear volume to determine if the tissue sample is representative of cancerous tissue.

System and method for automated stereology of cancer

An example method for diagnosing tumors in a subject by performing a quantitative analysis of a radiological image can include identifying a region of interest (ROI) in the radiological image, segmenting the ROI from the radiological image, identifying a tumor object in the segmented ROI and segmenting the tumor object from the segmented ROI. The method can also include extracting a plurality of quantitative features describing the segmented tumor object, and classifying the tumor object based on the extracted quantitative features. The quantitative features can include one or more texture-based features.

Systems and methods for diagnosing tumors in a subject by performing a quantitative analysis of texture-based features of a tumor object in a radiological image

This invention relates to a system and method for applying an ensemble of segmentations to microscopy images of a tissue sample to determine if the tissue sample is representative of cancerous tissue. The ensemble of segmentations is applied to a plurality of greyscale or color microscopy images to generate a final image level segmentation and a final blob level segmentation. The final image level segmentation and final blob level segmentation are used to calculate a mean nuclear volume to determine if the tissue sample is representative of cancerous tissue.

Systems and methods are provided for automatic determination of slice thickness of an image stack in a computerized stereology system, as well as automatic quantification of biological objects of interest within an identified slice of the image stack. Top and bottom boundaries of a slice can be identified by applying a thresholded focus function to determine just-out-of-focus focal planes. Objects within an identified slice can be quantified by performing a color processing segmentation followed by a gray-level processing segmentation. The two segmentation processes generate unique identifiers for features in an image that can then be used to produce a count of the features.

Growing community of inventors

Tampa, FL, United States of America

Lawrence O Hall