Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) are commonly used to assess patients with known or suspected pathologies of the lungs and liver. In particular, identification and quantification of possibly malignant regions identified in these high-resolution images is essential for accurate and timely diagnosis. However, careful quantitative assessment of lung and liver lesions is tedious and time consuming. This disclosure describes an automated end-to-end pipeline for accurate lesion detection and segmentation.

An imaging system includes a microscope to generate magnified images of regions of interest of a tissue sample, a camera to capture and store the magnified images, and a controller. The controller is configured to, for each magnification level in a sequence of increasing magnification levels, image one or more regions of interest of the tissue sample at the current magnification level. For each region of interest, data is generated defining one or more refined regions of interest based on the magnified image of the region of interest of the tissue sample at the current magnification level. Each refined region of interest corresponds to a proper subset of the tissue sample, and the refined regions of interest of the tissue sample provide the regions of interest to be imaged at a next magnification level from the sequence of increasing magnification levels.

An imaging system includes a microscope to generate magnified images of regions of interest of a tissue sample, a camera to capture and store the magnified images, and a controller. The controller is configured to, for each magnification level in a sequence of increasing magnification levels, image one or more regions of interest of the tissue sample at the current magnification level. For each region of interest, data is generated defining one or more refined regions of interest based on the magnified image of the region of interest of the tissue sample at the current magnification level. Each refined region of interest corresponds to a proper subset of the tissue sample, and the refined regions of interest of the tissue sample provide the regions of interest to be imaged at a next magnification level from the sequence of increasing magnification levels.

In implementations of segmenting objects in vector graphics images, an object segmentation system can obtain points that identify an object in a vector graphics image, and determine a region of interest in the image that includes the object based on the points that identify the object. The object segmentation system can generate a heat map from the points that identify the object in the image, and a rasterized region from rasterizing the region of interest. The object segmentation system can generate a mask from the rasterized region and the heat map, the mask identifying pixels of the object in the rasterized region, and determine, from the mask, paths of the vector graphics corresponding to the object.

In implementations of segmenting objects in vector graphics images, an object segmentation system can obtain points that identify an object in a vector graphics image, and determine a region of interest in the image that includes the object based on the points that identify the object. The object segmentation system can generate a heat map from the points that identify the object in the image, and a rasterized region from rasterizing the region of interest. The object segmentation system can generate a mask from the rasterized region and the heat map, the mask identifying pixels of the object in the rasterized region, and determine, from the mask, paths of the vector graphics corresponding to the object.

A method and instruction memory for processing acoustic images of a downhole casing to determine perforations of the tubular. The images may be acquired by an acoustic logging tool deployed into cased well. A Machine Learning model is trained to recognize regions of the acoustic images that are perforations or not, in order to calculate geometric properties of the perforation and overall casing. Renderings of the imaged casing may be overlaid with contours and properties of perforations to improve perforation, fracturing and producing operations.

A method and instruction memory for processing acoustic images of a downhole casing to determine perforations of the tubular. The images may be acquired by an acoustic logging tool deployed into cased well. A Machine Learning model is trained to recognize regions of the acoustic images that are perforations or not, in order to calculate geometric properties of the perforation and overall casing. Renderings of the imaged casing may be overlaid with contours and properties of perforations to improve perforation, fracturing and producing operations.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a segmentation neural network. In one aspect, a method comprises: obtaining data defining: (i) an image, and (ii) a respective class of each pixel in the image from a set of possible classes; determining a target segmentation of the image that comprises one or more target contrastive channels, wherein each target contrastive channel corresponds to a respective pair of classes including a respective first class and a respective second class from the set of possible classes; and training the segmentation neural network to process the image to generate a predicted segmentation that matches the target segmentation.

The present invention relates to a device (2) and a method (100) for determining at least one final two-dimensional image or slice for visualizing an object of interest in a three-dimensional ultrasound volume. The method (100) for determining at least one final two-dimensional image, the method comprises the steps: a) providing (101) a three-dimensional image of a body region of a patient body, wherein an applicator configured for fixating at least one radiation source is inserted into the body region; b) providing (102) an initial direction, in particular by randomly determining the initial direction within the three-dimensional image; c) repeating (103) the following sequence of steps s1) to s4): s1) determining (104), via a processing unit, a set-direction within the three-dimensional image based on the initial direction for the first sequence or based on a probability map determined during a previous sequence; s2) extracting (105), via the processing unit, an image-set of two-dimensional images from the three-dimensional image, such that the two-dimensional images of the image-set are arranged coaxially and subsequently in the set-direction; s3) applying (106), via the processing unit, an applicator pre-trained classification method to each of the two-dimensional images of the image-set resulting in a probability score for each of the two-dimensional images of the image-set indicating a probability of the applicator being depicted, in particular fully depicted, in the respective two-dimensional image of the image-set in a cross-sectional view; and s4) determining (107), via the processing unit, a probability-map representing the probability scores of the two-dimensional images of the image-set with respect to the set-direction; wherein the method comprises the further step: d) determining (108), via a processing unit and after finishing the last sequence, the two-dimensional image associated with the highest probability score, in particular from the image-set determined during the last sequence, as the final two-dimensional image. The invention provides an efficient way to ensure that the ultrasound volume has the required clinical information by providing the necessary scan planes having the object of interest e.g. the applicator (6) in a three-dimensional ultrasound volume.

The present invention relates to a device (2) and a method (100) for determining at least one final two-dimensional image or slice for visualizing an object of interest in a three-dimensional ultrasound volume. The method (100) for determining at least one final two-dimensional image, the method comprises the steps: a) providing (101) a three-dimensional image of a body region of a patient body, wherein an applicator configured for fixating at least one radiation source is inserted into the body region; b) providing (102) an initial direction, in particular by randomly determining the initial direction within the three-dimensional image; c) repeating (103) the following sequence of steps s1) to s4): s1) determining (104), via a processing unit, a set-direction within the three-dimensional image based on the initial direction for the first sequence or based on a probability map determined during a previous sequence; s2) extracting (105), via the processing unit, an image-set of two-dimensional images from the three-dimensional image, such that the two-dimensional images of the image-set are arranged coaxially and subsequently in the set-direction; s3) applying (106), via the processing unit, an applicator pre-trained classification method to each of the two-dimensional images of the image-set resulting in a probability score for each of the two-dimensional images of the image-set indicating a probability of the applicator being depicted, in particular fully depicted, in the respective two-dimensional image of the image-set in a cross-sectional view; and s4) determining (107), via the processing unit, a probability-map representing the probability scores of the two-dimensional images of the image-set with respect to the set-direction; wherein the method comprises the further step: d) determining (108), via a processing unit and after finishing the last sequence, the two-dimensional image associated with the highest probability score, in particular from the image-set determined during the last sequence, as the final two-dimensional image. The invention provides an efficient way to ensure that the ultrasound volume has the required clinical information by providing the necessary scan planes having the object of interest e.g. the applicator (6) in a three-dimensional ultrasound volume.