Abstract: Systems and techniques that facilitate data diversity visualization and/or quantification for machine learning models are provided. In various embodiments, a processor can access a first dataset and a second dataset, where a machine learning (ML) model is trained on the first dataset. In various instances, the processor can obtain a first set of latent activations generated by the ML model based on the first dataset, and a second set of latent activations generated by the ML model based on the second dataset. In various aspects, the processor can generate a first set of compressed data points based on the first set of latent activations, and a second set of compressed data points based on the second set of latent activations, via dimensionality reduction. In various instances, a diversity component can compute a diversity score based on the first set of compressed data points and second set of compressed data points.

Abstract: Systems/techniques that facilitate deep learning multi-planar reformatting of medical images are provided. In various embodiments, a system can access a three-dimensional medical image. In various aspects, the system can localize, via execution of a machine learning model, a set of landmarks depicted in the three-dimensional medical image, a set of principal anatomical planes depicted in the three-dimensional medical image, and a set of organs depicted in the three-dimensional medical image. In various instances, the system can determine an anatomical orientation exhibited by the three-dimensional medical image, based on the set of landmarks, the set of principal anatomical planes, or the set of organs. In various cases, the system can rotate the three-dimensional medical image, such that the anatomical orientation now matches a predetermined anatomical orientation.

Abstract: Techniques are described for generating high quality training data collections for training artificial intelligence (AI) models in the medical imaging domain. A method embodiment comprises receiving, by a system comprising processor, input indicating a clinical context associated with usage of a medical image dataset, and selecting, by the system, one or more data scrutiny metrics for filtering the medical image dataset based on the clinical context. The method further comprises applying, by the system, one or more image processing functions to the medical image dataset to generate metric values of the one or more data scrutiny metrics for respective medical images included in the medical image dataset, filtering, by the system, the medical image dataset into one or more subsets based on one or more acceptability criteria for the metric values.

Abstract: Systems and techniques that facilitate data diversity visualization and/or quantification for machine learning models are provided. In various embodiments, a processor can access a first dataset and a second dataset, where a machine learning (ML) model is trained on the first dataset. In various instances, the processor can obtain a first set of latent activations generated by the ML model based on the first dataset, and a second set of latent activations generated by the ML model based on the second dataset. In various aspects, the processor can generate a first set of compressed data points based on the first set of latent activations, and a second set of compressed data points based on the second set of latent activations, via dimensionality reduction. In various instances, a diversity component can compute a diversity score based on the first set of compressed data points and second set of compressed data points.

Abstract: The current disclosure provides systems and methods for improving a visualization of an image volume of a uterus and/or endometrium of a subject acquired using a transvaginal ultrasound system (TVUS). In one example, a method for the TVUS comprises extracting a medial axis of an endometrium of a received two-dimensional (2D) ultrasound image of a uterus of a subject; generating a uterine trace line based on the extracted medial axis; acquiring a three-dimensional (3D) image volume of the uterus based on the uterine trace line; and displaying the 3D image volume on a display device of the TVUS.

Abstract: Systems or techniques for facilitating learnable visual prompt engineering are provided. In various embodiments, a system can access a medical image and a pre-trained machine learning model that is configured to perform a diagnostic or prognostic inferencing task. In various aspects, the system can apply a pre-processing transformation to one or more pixels or voxels of the medical image, thereby yielding a transformed version of the medical image, wherein the pre-processing transformation can convert an input pixel or voxel intensity value to an output pixel or voxel intensity value via one or more parameters that are iteratively learned. In various instances, the system can perform the diagnostic or prognostic inferencing task, by executing the pre-trained machine learning model on the transformed version of the medical image.

Abstract: Systems/techniques that facilitate deep learning multi-planar reformatting of medical images are provided. In various embodiments, a system can access a three-dimensional medical image. In various aspects, the system can localize, via execution of a machine learning model, a set of landmarks depicted in the three-dimensional medical image, a set of principal anatomical planes depicted in the three-dimensional medical image, and a set of organs depicted in the three-dimensional medical image. In various instances, the system can determine an anatomical orientation exhibited by the three-dimensional medical image, based on the set of landmarks, the set of principal anatomical planes, or the set of organs. In various cases, the system can rotate the three-dimensional medical image, such that the anatomical orientation now matches a predetermined anatomical orientation.

Abstract: A method for performing one-shot anatomy localization includes obtaining a medical image of a subject. The method includes receiving a selection of both a template image and a region of interest within the template image, wherein the template image includes one or more anatomical landmarks assigned a respective anatomical label. The method includes inputting both the medical image and the template image into a trained vision transformer model. The method includes outputting from the trained vision transformer model both patch level features and image level features for both the medical image and the template image. The method still further includes interpolating pixel level features from the patch level features for both the medical image and the template image. The method includes utilizing the pixel level features within the region of interest of the template image to locate and label corresponding pixel level features in the medical image.

Abstract: A method for performing one-shot anatomy localization includes obtaining a medical image of a subject. The method includes receiving a selection of both a template image and a region of interest within the template image, wherein the template image includes one or more anatomical landmarks assigned a respective anatomical label. The method includes inputting both the medical image and the template image into a trained vision transformer model. The method includes outputting from the trained vision transformer model both patch level features and image level features for both the medical image and the template image. The method still further includes interpolating pixel level features from the patch level features for both the medical image and the template image. The method includes utilizing the pixel level features within the region of interest of the template image to locate and label corresponding pixel level features in the medical image.

Abstract: Systems or techniques that facilitate ensembled querying of example images via deep learning embeddings are provided. In various embodiments, a system can access a medical image associated with a medical patient. In various aspects, the system can generate an ensembled heat map indicating where in the medical image an anatomical structure is likely to be located, by executing an embedder neural network on the medical image and on a plurality of example medical images associated with other medical patients. In various instances, respective instantiations of the anatomical structure can be flagged in the plurality of example medical images by user-provided clicks.

Abstract: Systems and method for domain adaptation using pseudo-labelling and model certainty quantification for video data are provided. The method includes obtaining a source data and a target data each comprising a plurality of frames for processing by a machine learning module. The method comprises testing the target data to identify if a minimum number of frames exhibit a frame confidence score based on the source data and identifying salient region within the target data and measuring a degree of spatial consistency of the salient region over time.

Abstract: Systems/techniques that facilitate interpretable task-specific dimensionality-reduction are provided. In various embodiments, a system can access a three-dimensional medical image. In various aspects, the system can generate, via execution of a first deep learning neural network, a voxel-wise weight map corresponding to the three-dimensional medical image and a set of projection vectors corresponding to the three-dimensional medical image. In various instances, the system can generate a set of two-dimensional projection images of the three-dimensional medical image, based on the voxel-wise weight map and the set of projection vectors. In various cases, the first deep learning neural network can be trained in a serial pipeline with a second deep learning neural network that is configured to perform an inferencing task on two-dimensional inputs. This can cause the set of two-dimensional projection images to be tailored to the inferencing task.

Abstract: The current disclosure provides systems and methods for automatic image alignment of three-dimensional (3D) medical image volumes. The method includes pre-processing the 3D medical image volume by selecting a sub-volume of interest, detecting anatomical landmarks in the sub-volume using a deep neural network, estimating transformation parameters based on the anatomical landmarks to adjust rotation angles and translation of the sub-volume, adjusting the rotation angles and translation to produce a first aligned sub-volume, determining confidence in the transformation parameters based on the first aligned sub-volume, and iteratively refining the transformation parameters if the confidence is below a predetermined threshold. The disclosed approach for automated image alignment reduces the need for manual alignment and, increases a probability of the 3D image volume converging to a desired orientation compared to conventional approaches.

Abstract: Systems/techniques that facilitate explainable confidence estimation for landmark localization are provided. In various embodiments, a system can access a three-dimensional voxel array captured by a medical imaging scanner and can localize, via execution of a first deep learning neural network, a set of anatomical landmarks depicted in the three-dimensional voxel array. In various aspects, the system can generate a multi-tiered confidence score collection based on the set of anatomical landmarks and based on a training dataset on which the first deep learning neural network was trained. In various instances, the system can, in response to one or more confidence scores from the multi-tiered confidence score collection failing to satisfy a threshold, generate, via execution of a second deep learning neural network, a classification label that indicates an explanatory factor for why the one or more confidence scores failed to satisfy the threshold.

Abstract: Systems and methods for enhancing visualization and documentation of fibroid quantity, size, and location with respect to a uterus and endometrium in ultrasound imaging are provided. The method includes receiving, by at least one processor, an ultrasound volume including ultrasound image data of a region of interest that includes a uterus. The method includes automatically segmenting, by the at least one processor, the uterus and an endometrium in the ultrasound volume. The method includes segmenting one or more fibroids in the ultrasound image data. The method includes generating a classification of each of the one or more fibroids based on a location of each of the one or more fibroids with respect to the endometrium and the uterus. The method includes causing, by the at least one processor, a display system to present at least one image identifying the segmented uterus, endometrium, and one or more fibroids.

Abstract: Techniques are described for learning feature representations of medical images using a self-supervised learning paradigm and employing those feature representations for automating downstream tasks such as image retrieval, image classification and other medical image processing tasks. According to an embodiment, computer-implemented method comprises generating alternate view images for respective medical images included in set of training images using one or more image augmentation techniques or one or more image selection techniques tailored based on domain knowledge associated with the respective medical images. The method further comprises training a transformer network to learn reference feature representations for the respective medical images using their alternate view images and a self-supervised training process.

Abstract: Techniques are described for refining or updating medical image inferencing models post deployment using synthetic images generated from non-image data feedback. In an example, a system can comprise a memory that stores computer-executable components and a processor that executes the computer-executable components stored in the memory. The computer-executable components can comprise an image generation component that generates synthetic medical images based on feedback information associated with performance of a medical image inferencing model received in association with application of the medical image inferencing model to medical images in a deployment environment, wherein the feedback information excludes image data. The computer-executable components can further comprise a refinement component that updates the medical image inferencing model using the synthetic images and a model updating process.

Abstract: Systems/techniques that facilitate data candidate querying via embeddings for deep learning refinement are provided. In various embodiments, a system can access a test data candidate provided by a client, generate, via a first deep learning neural network, an inferencing output based on the test data candidate, and access feedback indicating whether the client accepts or rejects the inferencing output. In various aspects, the system can generate, via at least one second deep learning neural network, at least one embedding based on the test data candidate. In various instances, the system can, in response to the feedback indicating that the client rejects the inferencing output, identify, in a candidate-embedding dataset, one or more data candidates whose embeddings are within a threshold level of similarity to the at least one embedding and can retrain the first deep learning neural network based on the one or more data candidates.

Abstract: A medical imaging system includes at least one medical imaging device providing image data of a subject and a processing system programmed to generate a plurality of training images having simulated medical conditions by blending a pathology region from a plurality of template source images to a plurality of target images. The processing system is further programmed to train a deep learning network model using the plurality of training images and input the image data of the subject to the deep learning network model. The processing system is further programmed to generate a medical image of the subject based on the output of the deep learning network model.

Abstract: Various methods and ultrasound imaging systems are provided for segmenting an object. In one example, a method includes accessing a volumetric ultrasound dataset, receiving an identification of a seed point for an object in an image generated based on the volumetric ultrasound dataset, and implementing a two-dimensional segmentation model on a first plurality of parallel slices based on the seed point to generate a first plurality of segmented regions. The method includes implementing the two-dimensional segmentation model on a second plurality of parallel slices based on the seed point to generate a second plurality of segmented regions. The method includes generating a detected region by accumulating the first plurality of segmented regions and the second plurality of segmented regions. The method includes implementing a shape completion model to generate a three-dimensional shape model for the object, and displaying rendering of the object based on the three-dimensional shape model.