Abstract: Knowledge distillation method for fracture detection includes obtaining medical images including region-level labeled images, image-level diagnostic positive images, and image-level diagnostic negative images, in chest X-rays; performing a supervised pre-training process on the region-level labeled images and the image-level diagnostic negative images to train a neural network to generate pre-trained weights; and performing a semi-supervised training process on the image-level diagnostic positive images using the pre-trained weights. A teacher model is employed to produce pseudo ground-truths (GTs) on the image-level diagnostic positive images for supervising training of a student model, and the pseudo GTs are processed by an adaptive asymmetric label sharpening (AALS) operator to produce sharpened pseudo GTs to provide positive detection responses on the image-level diagnostic positive images.

Abstract: Methods and systems for image registration using an intelligent artificial agent are disclosed. In an intelligent artificial agent based registration method, a current state observation of an artificial agent is determined based on the medical images to be registered and current transformation parameters. Action-values are calculated for a plurality of actions available to the artificial agent based on the current state observation using a machine learning based model, such as a trained deep neural network (DNN). The actions correspond to predetermined adjustments of the transformation parameters. An action having a highest action-value is selected from the plurality of actions and the transformation parameters are adjusted by the predetermined adjustment corresponding to the selected action. The determining, calculating, and selecting steps are repeated for a plurality of iterations, and the medical images are registered using final transformation parameters resulting from the plurality of iterations.

Abstract: Methods and systems for image registration using an intelligent artificial agent are disclosed. In an intelligent artificial agent based registration method, a current state observation of an artificial agent is determined based on the medical images to be registered and current transformation parameters. Action-values are calculated for a plurality of actions available to the artificial agent based on the current state observation using a machine learning based model, such as a trained deep neural network (DNN). The actions correspond to predetermined adjustments of the transformation parameters. An action having a highest action-value is selected from the plurality of actions and the transformation parameters are adjusted by the predetermined adjustment corresponding to the selected action. The determining, calculating, and selecting steps are repeated for a plurality of iterations, and the medical images are registered using final transformation parameters resulting from the plurality of iterations.

Abstract: The present disclosure provides a computer-implemented method, a device, and a computer program product for radiographic bone mineral density (BMD) estimation. The method includes receiving a plain radiograph, detecting landmarks for a bone structure included in the plain radiograph, extracting an ROI from the plain radiograph based on the detected landmarks, estimating the BMD for the ROI extracted from the plain radiograph by using a deep neural network.

Abstract: A vertebra localization and identification method includes: receiving one or more images of vertebrae of a spine; applying a machine learning model on the one or more images to generate three-dimensional (3-D) vertebra activation maps of detected vertebra centers; performing a spine rectification process on the 3-D vertebra activation maps to convert each 3-D vertebra activation map into a corresponding one-dimensional (1-D) vertebra activation signal; performing an anatomically-constrained optimization process on each 1-D vertebra activation signal to localize and identify each vertebra center in the one or more images; and outputting the one or more images, wherein on each of the one or more outputted images, a location and an identification of each vertebra center are specified.

Abstract: A method and system for deep learning based cardiac electrophysiological model personalization is disclosed. Electrophysiological measurements of a patient, such as an ECG trace, are received. A computational cardiac electrophysiology model is personalized by calculating patient-specific values for a parameter of the computational cardiac electrophysiology model based at least on the electrophysiological measurements of the patient using a trained deep neural network (DNN). The parameter of the computational cardiac electrophysiology model corresponds to a spatially varying electrical cardiac tissue property.

Abstract: Methods and systems for image registration using an intelligent artificial agent are disclosed. In an intelligent artificial agent based registration method, a current state observation of an artificial agent is determined based on the medical images to be registered and current transformation parameters. Action-values are calculated for a plurality of actions available to the artificial agent based on the current state observation using a machine learning based model, such as a trained deep neural network (DNN). The actions correspond to predetermined adjustments of the transformation parameters. An action having a highest action-value is selected from the plurality of actions and the transformation parameters are adjusted by the predetermined adjustment corresponding to the selected action. The determining, calculating, and selecting steps are repeated for a plurality of iterations, and the medical images are registered using final transformation parameters resulting from the plurality of iterations.

Abstract: The present disclosure provides a method, a device, and a computer program product using a self-supervised anatomical embedding (SAM) method.

Abstract: An image segmentation method includes generating a CTN (contour transformer network) model for image segmentation, where generating the CTN model includes providing an annotated image, the annotated image including an annotated contour, providing a plurality of unannotated images, pairing the annotated image to each of the plurality of unannotated images to obtain a plurality of image pairs, feeding the plurality of image pairs to an image encoder to obtain a plurality of first-processed image pairs, and feeding the plurality of first-processed image pairs to a contour tuner to obtain a plurality of second-processed image pairs.

Abstract: Methods and systems for image registration using an intelligent artificial agent are disclosed. In an intelligent artificial agent based registration method, a current state observation of an artificial agent is determined based on the medical images to be registered and current transformation parameters. Action-values are calculated for a plurality of actions available to the artificial agent based on the current state observation using a machine learning based model, such as a trained deep neural network (DNN). The actions correspond to predetermined adjustments of the transformation parameters. An action having a highest action-value is selected from the plurality of actions and the transformation parameters are adjusted by the predetermined adjustment corresponding to the selected action. The determining, calculating, and selecting steps are repeated for a plurality of iterations, and the medical images are registered using final transformation parameters resulting from the plurality of iterations.

Abstract: A method of opportunistic screening of osteoporosis includes obtaining a plain film chest X-ray (CXR); extracting regions of interest (ROIs) from the plain film CXR; and providing individual bone mineral density (BMD) scores corresponding to the extracted ROIs and a joint BMD corresponding to the plain film CXR based on a multi-ROI model by performing: inputting the extracted ROIs into a backbone network to generate individual feature vectors, each individual feature vector corresponding to one of the extracted ROIs; concatenating the individual feature vectors into a joint feature vector; individually decoding the individual feature vectors by a shared fully connected (FC) layer to generate the individual BMDs, each individual BMD corresponding to one of the individual feature vectors; and decoding the joint feature vector by a separate FC layer to generate the joint BMD.

Abstract: A method for estimating bone mineral density (BMD) includes obtaining an image and cropping one or more regions-of-interest (ROIs) in the image, taking the one or more ROIs as input to a network model for estimating BMDs, training the network model on the labeled one or more ROIs with one or more loss functions to obtain a pre-trained model in a supervised pre-training stage, and fine-tuning the pre-trained model on a first plurality of data representing the labeled one or more ROIs and a second plurality of data representing unlabeled region to determine a fine-tuned network model for estimating BMDs in a semi-supervised self-training stage. The one or more loss functions includes a specific adaptive triplet loss (ATL) configured to encourage distances between one or more feature embedding vectors correlated to differences among the BMDs.

Abstract: For non-invasive EP mapping, a sparse number of electrodes (e.g., 10 in a typical 12-lead ECG exam setting) are used to generate an EP map without requiring preoperative 3D image data (e.g. MR or CT). An imager (e.g., a depth camera) captures the surface of the patient and may be used to localize electrodes in any positioning on the patient. Two-dimensional (2D) x-rays, which are commonly available, and the surface of the patient are used to segment the heart of the patient. The EP map is then generated from the surface, heart segmentation, and measurements from the electrodes.

Abstract: For registration of medical images with deep learning, a neural network is designed to include a diffeomorphic layer in the architecture. The network may be trained using supervised or unsupervised approaches. By enforcing the diffeomorphic characteristic in the architecture of the network, the training of the network and application of the learned network may provide for more regularized and realistic registration.

Abstract: Knowledge distillation method for fracture detection includes obtaining medical images including region-level labeled images, image-level diagnostic positive images, and image-level diagnostic negative images, in chest X-rays; performing a supervised pre-training process on the region-level labeled images and the image-level diagnostic negative images to train a neural network to generate pre-trained weights; and performing a semi-supervised training process on the image-level diagnostic positive images using the pre-trained weights. A teacher model is employed to produce pseudo ground-truths (GTs) on the image-level diagnostic positive images for supervising training of a student model, and the pseudo GTs are processed by an adaptive asymmetric label sharpening (AALS) operator to produce sharpened pseudo GTs to provide positive detection responses on the image-level diagnostic positive images.

Abstract: The present disclosure provides a method, a device, and a computer program product using a self-supervised anatomical embedding (SAM) method.

Abstract: A method and system for 3D/3D medical image registration. A digitally reconstructed radiograph (DRR) is rendered from a 3D medical volume based on current transformation parameters. A trained multi-agent deep neural network (DNN) is applied to a plurality of regions of interest (ROIs) in the DRR and a 2D medical image. The trained multi-agent DNN applies a respective agent to each ROI to calculate a respective set of action-values from each ROI. A maximum action-value and a proposed action associated with the maximum action value are determined for each agent. A subset of agents is selected based on the maximum action-values determined for the agents. The proposed actions determined for the selected subset of agents are aggregated to determine an optimal adjustment to the transformation parameters and the transformation parameters are adjusted by the determined optimal adjustment.

Abstract: A vertebra localization and identification method includes: receiving one or more images of vertebrae of a spine; applying a machine learning model on the one or more images to generate three-dimensional (3-D) vertebra activation maps of detected vertebra centers; performing a spine rectification process on the 3-D vertebra activation maps to convert each 3-D vertebra activation map into a corresponding one-dimensional (1-D) vertebra activation signal; performing an anatomically-constrained optimization process on each 1-D vertebra activation signal to localize and identify each vertebra center in the one or more images; and outputting the one or more images, wherein on each of the one or more outputted images, a location and an identification of each vertebra center are specified.

Abstract: A method for performing computer-aided diagnosis (CAD) based on a medical scan image includes: pre-processing the medical scan image to produce an input image, a flipped image, and a spatial alignment transformation corresponding to the input image and the flipped image; performing Siamese encoding on the input image to produce an encoded input feature map; performing Siamese encoding on the flipped image to produce an encoded flipped feature map; performing a feature alignment using the spatial alignment transformation on the encoded flipped feature map to produce an encoded symmetric feature map; and processing the encoded input feature map and the encoded symmetric feature map to generate a diagnostic result indicating presence and locations of anatomical abnormalities in the medical scan image.

Abstract: Systems and methods are provided for performing medical imaging analysis. Input medical imaging data is received for performing a particular one of a plurality of medical imaging analyses. An output that provides a result of the particular medical imaging analysis on the input medical imaging data is generated using a neural network trained to perform the plurality of medical imaging analyses. The neural network is trained by learning one or more weights associated with the particular medical imaging analysis using one or more weights associated with a different one of the plurality of medical imaging analyses. The generated output is outputted for performing the particular medical imaging analysis.