Abstract: Systems and methods are provided for optimizing a deep learning model. A multi-site dataset associated with different clinical sites and a deployment dataset associated with a deployment clinical site are received. A deep learning model is trained based on the multi-site dataset. The trained deep learning model is optimized based on the deployment dataset. The optimized trained deep learning model is output.

Abstract: Systems and methods are provided for classifying an abnormality in a medical image. An input medical image depicting a lesion is received. The lesion is localized in the input medical image using a trained localization network to generate a localization map. The lesion is classified based on the input medical image and the localization map using a trained classification network. The classification of the lesion is output. The trained localization network and the trained classification network are jointly trained.

Abstract: Systems and methods for reducing false positive detections of malignant lesions are provided. A candidate malignant lesion is detected in one or more medical images, such as, e.g., multi-parametric magnetic resonance images. One or more patches associated with the candidate malignant lesion are extracted from the one or more medical images. The candidate malignant lesion is classified as being a true positive detection of a malignant lesion or a false positive detection of the malignant lesion based on the one or more extract patches using a trained machine learning network. The results of the classification are output.

Abstract: In order to reduce computation time and provide more accurate solutions for bi-directional, multi-modal image registration, a learning-based unsupervised multi-modal deformable image registration method that does not require any aligned image pairs or anatomical landmarks is provided. A bi-directional registration function is learned based on disentangled shape representation by optimizing a similarity criterion defined on both latent space and image space.

Abstract: Systems and methods for computing a transformation for correction motion between a first medical image and a second medical image are provided. One or more landmarks are detected in the first medical image and the second medical image. A first tree of the anatomical structure is generated from the first medical image based on the one or more landmarks detected in the first medical image and a second tree of the anatomical structure is generated from the second medical image based on the one or more landmarks detected in the second medical image. The one or more landmarks detected in the first medical image are mapped to the one or more landmarks detected in the second medical image based on the first tree and the second tree. A transformation to align the first medical image and the second medical image is computed based on the mapping.

Abstract: In an method for training artificial intelligence entities (AIE) for abnormality detection, medical imaging data of the human organ is provided as training data having training samples, the medical imaging data including imaging results from different types of imaging techniques for each training sample of the training data, a pre-trained or randomly initialized AIE is provided, and the AIE is trained using the provided training samples. The training may include, for at least one training sample, a first loss function for a sub-structure of the AIE is calculated independently of a first spatial region of the human organ, and, for a training sample, a second loss function for a sub-structure of the AIE is calculated independently of a second spatial region of the human organ. The AIE may be trained using the calculated first loss function and the calculated second loss function.

Abstract: Systems and methods are provided for classifying an abnormality in a medical image. An input medical image depicting a lesion is received. The lesion is localized in the input medical image using a trained localization network to generate a localization map. The lesion is classified based on the input medical image and the localization map using a trained classification network. The classification of the lesion is output. The trained localization network and the trained classification network are jointly trained.

Abstract: Systems and methods for computing a transformation for correction motion between a first medical image and a second medical image are provided. One or more landmarks are detected in the first medical image and the second medical image. A first tree of the anatomical structure is generated from the first medical image based on the one or more landmarks detected in the first medical image and a second tree of the anatomical structure is generated from the second medical image based on the one or more landmarks detected in the second medical image. The one or more landmarks detected in the first medical image are mapped to the one or more landmarks detected in the second medical image based on the first tree and the second tree. A transformation to align the first medical image and the second medical image is computed based on the mapping.

Abstract: Systems and methods for reducing false positive detections of malignant lesions are provided. A candidate malignant lesion is detected in one or more medical images, such as, e.g., multi-parametric magnetic resonance images. One or more patches associated with the candidate malignant lesion are extracted from the one or more medical images. The candidate malignant lesion is classified as being a true positive detection of a malignant lesion or a false positive detection of the malignant lesion based on the one or more extract patches using a trained machine learning network. The results of the classification are output.

Abstract: In order to reduce computation time and provide more accurate solutions for bi-directional, multi-modal image registration, a learning-based unsupervised multi-modal deformable image registration method that does not require any aligned image pairs or anatomical landmarks is provided. A bi-directional registration function is learned based on disentangled shape representation by optimizing a similarity criterion defined on both latent space and image space.