Abstract: Disclosed are techniques for building a dynamic dictionary and using the dictionary to remove phrases or words appearing in and out of context in a document. The techniques include, for example, receiving electronic health record (EHR) data, determining, using natural language processing (NLP), an instance of a personal health information (PHI) phrase in the EHR data based on a NLP system confidence metric being above a threshold, determining another instance of the PHI phrase in the EHR data that does not have the same context as the first context, removing the instances of the PHI phrase from the EHR data to produce cleaned EHR data, and taking an action based on the cleaned EHR data. The confidence metric can indicate likelihood that the PHI phrase is a PHI phrase and the metric can be based at least in part on a first context of the PHI phrase.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for determining a subset of brain data of a patient. One of the methods includes obtaining data characterizing a brain of a patient; determining a first prompt for presentation to a user; obtaining a first user input characterizing a first response to the first prompt; determining, using the first response to the first prompt, a second prompt for presentation to the user; obtaining a second user input characterizing a second response to the second prompt, wherein at least one of the first prompt or the second prompt seek a response based on a clinical observation of the patient; and determining a subset of the obtained data using the first response to the first prompt and the second response to the second prompt.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating explainability data that explains a medical condition in a subject. In one aspect, a method comprises: obtaining data identifying a plurality of brain parcels that are predicted to be relevant to the medical condition; receiving fMRI data for a brain of a subject; processing the fMRI data for the brain of the subject to determine a respective activation score for each of the plurality of brain parcels that are predicted to be relevant to the medical condition; determining, for each of the plurality of brain parcels that are predicted to be relevant to the medical condition, a relative activation score for the brain parcel; and taking an action based on the relative activation scores.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for determining a subset of brain data of a patient. One of the methods comprises providing connectivity data for presentation to a user, the connectivity data characterizing, for each pair of parcellations comprising a first parcellation and a second parcellation from a plurality of parcellations, a degree of correlation between the brain activity of the first parcellation and the brain activity of the second parcellation in the brain of a patient; determining one or more elements of interest in the connectivity data; determining one or more parcellations associated with elements of interest in the connectivity data; obtaining brain atlas data; determining a subset of the brain atlas data associated with the determined parcellations; and providing the subset of the brain atlas data to a user device for rendering the subset to the user.

Abstract: Disclosed herein are systems and methods for interactive graphical user interfaces (GUIs) that users (e.g., medical professionals) can use to interact with modelled versions of brains and easily and intuitively analyze deep and/or lateral structures in the brain. A user can, for example, selectively view structures and their connectivity data (e.g., nodes and edges) relative to other structures and connectivity data over a representation of a particular patient's brain. Emphasis can be minimized for certain foreground nodes and edges (e.g., lateral structures) to make it easier for the user to focus on and analyze deeper structures that otherwise can be challenging to visualize and understand. A method can include overlaying deep and non-deep nodes on a representation of a brain, displaying the representation of the brain in a GUI, receiving user input indicating interest in focusing on one or more deep nodes, and taking an action based on the input.

Abstract: Methods and systems for measuring 3-dimensional distances on medical images are described. One example method includes receiving at least a first 2-dimensional image on a first plane and a second 2-dimensional image on a second plane, wherein the first plane and the second plane are different planes, selecting a first point on the first 2-dimensional image as a start point, selecting a second point on the second 2-dimensional image as an end point, determining a shortest path between the start point and the end point, performing a measurement of the shortest path between the start point and the end point, and, taking an action based on the measurement.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating visualization data for a selected neuro-network. In one aspect, a method comprises: receiving selection data selecting a network in a brain of a subject; processing magnetic resonance image data of the brain to identify a set of tracts that are predicted to be included in the selected network; processing the set of tracts to identify a proper subset of the set of tracts as being spurious tracts; generating a set of valid tracts by filtering the spurious tracts from the set of tracts that are predicted to be included in the selected network; providing visualization data for the selected network showing a three-dimensional spatial representation of both: (i) the valid tracts, and (ii) the spurious tracts, wherein the spurious tracts are visually distinguished from the valid tracts.

Abstract: A pre-event connectome of a subject brain is accessed, the pre-event connectome defining i) first functional nodes in the subject brain and ii) first edges that represent connections between the first functional nodes before the subject has undergone an event. A post-event connectome of the subject brain is accessed, the post-event connectome defining i) second functional nodes in the subject brain and ii) second edges that represent connections between the second functional nodes after the subject has undergone the event. A connectome-difference map data is generated that records the difference between the pre-event connectome and the post-event connectome. An action is taken based on the connectome-difference map data.

Abstract: fMRI data of a subject brain is accessed and may include a plurality of time-sequenced volumetric images of activity in a subject brain. A plurality of emotion vectors are accessed, each emotion vector tagged with a specified emotional state. From the fMRI data and using the emotion vectors, a plurality of fMRI state vectors are determined at various points in time, where each fMRI state vector is a combination of the emotion vectors and represents the state of the subject brain at a particular point in time. Flow data is determined to identify a trajectory, over time, of the subject brain as reflected by the fMRI data, through a state space defined by the emotion vectors, where the flow data is based at least in part on the fMRI vectors at various points in time. From the flow data, data is generated that shows changes, through time, in at least one emotional state of the brain.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for predicted brain data of a patient. One of the methods includes receiving montage configuration data for a specified montage; receiving raw EEG data captured using the specified montage from a brain of a particular subject; generating, using the montage configuration data and the raw EEG data, EEG connectivity data for the specified montage; using a generative neural network to map the EEG connectivity data to predicted fMRI connectivity data, the generative neural network having been trained using training EEG-fMRI connectivity data pairs, each pair comprising EEG connectivity data of a subject and fMRI connectivity data of the same subject; and taking an action based on the predicted fMRI connectivity data.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing brain data using autoencoder neural networks. One of the methods includes obtaining brain data captured by one or more sensors characterizing brain activity of a patient; processing the brain data to generate modified brain data that characterizes a predicted local effect of a future treatment on the brain of the patient; processing the modified brain data using an autoencoder neural network to generate reconstructed brain data; and determining, using the reconstructed brain data, a predicted global effect of the future treatment on the brain of the patient.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating explainability data that explains a medical condition in a subject. In one aspect, a method comprises: obtaining data identifying a plurality of brain parcels that are predicted to be relevant to the medical condition; receiving fMRI data for a brain of a subject; processing the fMRI data for the brain of the subject to determine a respective activation score for each of the plurality of brain parcels that are predicted to be relevant to the medical condition; determining, for each of the plurality of brain parcels that are predicted to be relevant to the medical condition, a relative activation score for the brain parcel; and taking an action based on the relative activation scores.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for harmonizing diffusion tensor images. One of the methods includes obtaining a diffusion tensor image; determining a set of RISH features for the diffusion tensor image; processing a model input generated from the set of RISH features using a machine learning model to generate a model output identifying an image transformation from a set of image transformations, wherein each image transformation in the set of image transformations corresponds to a respective different first MRI scanner and represents a transformation that, when applied to first diffusion tensor images captured by the first MRI scanner, harmonizes the first diffusion tensor images with second diffusion tensor images captured by a reference MRI scanner; and processing the diffusion tensor image using the identified image transformation to generate a harmonized diffusion tensor image.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for filtering a set of tracts that are predicted to be included in a selected neuro-network. In one aspect, a method comprises: receiving selection data selecting a network of a brain, processing magnetic resonance image data of the brain to identify a set of tracts that are predicted to be included in the selected network, determining a blocking surface for the selected network, comprising: obtaining a set of parcellations for the selected network and determining, based on a respective position in the brain of each parcellation in the set of parcellations, a set of parameters defining the blocking surface, and generating filtered tract data by filtering the set of tracts that are predicted to be included in the selected network to remove tracts that intersect the blocking surface.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a mental health prediction for a patient using a centrality ranking of the brain of the patient. One of the methods includes obtaining brain data of a patient, wherein the brain data comprises, for each of a plurality of pairs of parcellations formed from a set of parcellations where each pair comprises a first parcellation and a second parcellation, data characterizing a number of tracts connecting the first parcellation and the second parcellation; determining a network graph from the brain data; generating, for each of a plurality of nodes in the network graph, a measure of centrality of the node; determining a centrality ranking of the plurality of nodes of the network graph according to the respective measures of centrality; generating a mental health prediction for the patient using the determined centrality ranking.

Abstract: Accessed from memory are i) a first brain image, ii) a second brain image, iii) a difference map, and iv) a connectivity element. A GUI may include: concurrently displaying each of these elements. User input to the GUI is received. A location of the user input is determined. The operations also include responsive to determining that the location of the user input is within the display of the first brain image, modifications are made to the display of the i) second brain image, ii) the difference map, and iii) the connectivity element to highlight elements of the display that correspond to an element of the first brain image that corresponds to the location of the user input.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for displaying hubs in brain data. One of the methods includes receiving hub data, the hub data comprising anatomical data and connectivity data of at least part of a brain, the connectivity data comprising a first plurality of nodes and a second plurality of edges, each edge of the second plurality of edges connecting a pair of nodes among the first plurality of nodes; forwarding the hub data to a user computer for display; receiving a first user indication requesting a display of unexpected hub data; determining, in response to the first user indication, unexpected hub data comprising nodes that have connectivity outside of a specified threshold; and forwarding the unexpected hub data to the user computer for display.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a mental health prediction for a patient. One of the methods includes obtaining brain data captured by one or more sensors characterizing a brain of a patient; determining a network graph from the brain data, wherein: a node of the network graph corresponds to a parcellation in the brain of the patient, and a subset of nodes of the network graph corresponds to a particular functional area of the brain; generating, for each of a plurality of nodes in the network graph, a measure of centrality of the node; generating, for the particular functional area of the brain and using the generated measures of centrality, a health score representing a measure of health of the functional area of the brain; generating a mental health prediction for the patient using the health score.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying one or more invalid images characterizing the brain of a patient. One of the methods includes obtaining a data object comprising one or more images characterizing a brain of a patient; processing the data object to determine one or more particular images that are invalid; and providing, for display to a user on a graphical interface, data characterizing i) each particular image and ii) a respective reason that each particular image is invalid selected from a set of possible reasons that an image may be invalid.

Abstract: DICOM data is automatically prepared for transit outside of the clinical-data infrastructure, by examining a plurality of metadata fields in the corresponding metadata in the DICOM data; identifying a first subset of the metadata fields as containing private information; identifying a second subset of the metadata fields as private-information free; accessing at least some of the plurality of layers of the DICOM data; and transforming the accessed layers into a single transmission-image, the transmission-image being in a format i) other than DICOM and ii) that stores the second subset of the metadata fields as transmission-metadata in a scheme that is non-redundant for a given transmission-image.