Abstract: A natural language understanding method begins with a radiological report text containing clinical findings. Errors in the text are corrected by analyzing character-level optical transformation costs weighted by a frequency analysis over a corpus corresponding to the report text. For each word within the report text, a word embedding is obtained, character-level embeddings are determined, and the word and character-level embeddings are concatenated to a neural network which generates a plurality of NER tagged spans for the report text. A set of linked relationships are calculated for the NER tagged spans by generating masked text sequences based on the report text and determined pairs of potentially linked NER spans. A dense adjacency matrix is calculated based on attention weights obtained from providing the one or more masked text sequences to a Transformer deep learning network, and graph convolutions are then performed over the calculated dense adjacency matrix.

Abstract: For training data pairs comprising training text (a radiological report) and training images (radiological images associated with the radiological report), a first encoder network determines word embeddings for the training text. A concept is generated from the operation of layers of the first encoder network, which is regularized by a first loss between the generated concept and a labeled concept for the training text. A second encoder network determines features for the training image. A heatmap is generated from the operation of layers of the second encoder network, which is regularized by a second loss between the generated heatmap and a labeled heatmap for the training image. A categorical cross entropy loss is calculated between a diagnostic quality category (classified by an error encoder) and a labeled diagnostic quality category for the training data pair. A total loss function comprising the first, second, and categorical cross entropy losses is minimized.

Abstract: A database merger method (20) merges two or more anonymized healthcare databases (X, Y). Each anonymized healthcare database has personally identifying information anonymized including having medical care units replaced by medical care unit placeholders. In the database merger method, statistical patient feature distributions are computed for medical care unit placeholders in the anonymized healthcare databases. Medical care unit placeholders in different anonymized healthcare databases are matched by matching corresponding statistical patient feature distributions for the respective medical care unit placeholders. Patients in different anonymized healthcare databases are matched. The patient matching is performed within matched pairs of medical care unit placeholders to improve computational efficiency.

Abstract: For training data pairs comprising training text (a radiological report) and training images (radiological images associated with the radiological report), a first encoder network determines word embeddings for the training text. A concept is generated from the operation of layers of the first encoder network, which is regularized by a first loss between the generated concept and a labeled concept for the training text. A second encoder network determines features for the training image. A heatmap is generated from the operation of layers of the second encoder network, which is regularized by a second loss between the generated heatmap and a labeled heatmap for the training image. A categorical cross entropy loss is calculated between a diagnostic quality category (classified by an error encoder) and a labeled diagnostic quality category for the training data pair. A total loss function comprising the first, second, and categorical cross entropy losses is minimized.

Abstract: A natural language understanding method begins with a radiological report text containing clinical findings. Errors in the text are corrected by analyzing character-level optical transformation costs weighted by a frequency analysis over a corpus corresponding to the report text. For each word within the report text, a word embedding is obtained, character-level embeddings are determined, and the word and character-level embeddings are concatenated to a neural network which generates a plurality of NER tagged spans for the report text. A set of linked relationships are calculated for the NER tagged spans by generating masked text sequences based on the report text and determined pairs of potentially linked NER spans. A dense adjacency matrix is calculated based on attention weights obtained from providing the one or more masked text sequences to a Transformer deep learning network, and graph convolutions are then performed over the calculated dense adjacency matrix.

Abstract: A device, system, and method optimizes a healthcare network. The method is performed at a device of a healthcare organization, the healthcare organization having a healthcare network including a plurality of healthcare providers. The method includes selecting a healthcare provider to be evaluated. The method includes determining a score for the healthcare provider, the score being based upon at least one of first information relative to the healthcare provider and second information relative to the healthcare network. The method includes generating a recommendation for the healthcare provider based upon the score.

Abstract: A health care performance assessment apparatus (10) includes at least one processor (16, 20, 22, 24) programmed to: collect information associated with medically-related events from a plurality of databases; group the collected information associated with medically-related events into episode of care (EOC) data structures wherein each EOC data structure contains the collected information for a single EOC treating a medical condition or providing a medical procedure to a single patient; store the EOC data structures in an EOC repository (18); group EOC data structures having a chosen commonality into at least one cohort; and calculate at least one key performance indicators (KPIs) for the cohorts. A display (2) is configured to display the KPIs for at least one selected cohort.

Abstract: An electronic processor (14) is programmed to perform integration (16) of N anonymized healthcare databases (10). For in a pair of databases (i,j) of the N anonymized healthcare databases, a set of features is identified (44) each contained in both databases i and j of the pair of databases (i,j). A conversion table is generated (46, 48) that matches patients of the pair of databases based on patient similarity measured by the set of features. The identifying and generating operations are repeated (50) for each unique pair of databases of the N anonymized healthcare databases to generate N(N 1)/2 conversion tables (20). The electronic processor is further programmed to perform a patient data retrieval process (18) which receives a patient ID of a patient in one of the N anonymized healthcare databases and retrieves patient data for the patient contained in the N anonymized healthcare databases using the N(N?1)/2 conversion tables.

Abstract: A device, system, and method optimizes a patient flow. The method is performed at a device of a healthcare organization, the healthcare organization having a healthcare network including a plurality of healthcare providers. The method includes determining a step in a patient flow for a patient of a primary care physician (PCP) associated with the healthcare network based upon first information relative to the patient. The method includes determining a referral of a healthcare provider to perform the step based upon the first information and second information relative to a region associated with the patient and the healthcare organization. The method includes determining whether the referral is acceptable based upon third information relative to the healthcare provider and the healthcare organization. The method includes generating a recommendation including the referral for the PCP when the referral is acceptable.

Abstract: A database merger method (20) merges two or more anonymized healthcare databases (X, Y). Each anonymized healthcare database has personally identifying information anonymized including having medical care units replaced by medical care unit placeholders. In the database merger method, statistical patient feature distributions are computed for medical care unit placeholders in the anonymized healthcare databases. Medical care unit placeholders in different anonymized healthcare databases are matched by matching corresponding statistical patient feature distributions for the respective medical care unit placeholders. Patients in different anonymized healthcare databases are matched. The patient matching is performed within matched pairs of medical care unit placeholders to improve computational efficiency.

Abstract: A system for uniformly correlating unstructured entry features included in unstructured therapy entries to associated therapy features in structured therapy information is disclosed. The system obtains unstructured therapy entries and identifies unstructured entry features within the individual unstructured therapy entries. The unstructured therapy entry features are correlated to corresponding associated therapy features. The correlation of unstructured entry features to associated therapy features is based on contextual information associated with the individual unstructured therapy entries. Contextual information associated with the unstructured therapy entry includes the syntax of the unstructured therapy entry, a creator of the unstructured therapy entry, and/or the format of the unstructured therapy entry.

Abstract: Data analysis of altered data includes analyzing (64) a test data set (14) with a data analysis technique using one or more configured processors (30) which create one or more analytical measures, and the test data set selected from an altered data set (12) according to a confidence score. At least one reliability measure of the one or more analytical measure is calculated using the configured one or more processors based on similarity of the one or more analytical measures and same analytic measures created from the data analysis technique applied to one or more reliability test data sets (16, 18) selected from the altered data set according to different confidence scores.

Abstract: A risk-adjusted assessment of a target facility's quality measures (e.g. mortality rate, length of stay, readmission rate, complications rate, etc.) is determined with respect to the quality measures of a broader population base. Patient cohorts are identified corresponding to particular ailments or treatments, and the target facility's risk-adjusted quality measures are determined for each cohort. When a particular quality measure for a target cohort indicates poor performance, factors that are determined to be relevant to the patients' outcomes are identified and used to create a control group of patients in the broader population who exhibit similar factors but had better outcomes than the patients of the target cohort. The care process (treatments, medications, interventions, etc.) that each of the target patients received is compared to the care process that each of the control patients received, to identify potential root-causes of the poorer performance.

Abstract: A method includes retrieving de-identified records for individuals from at least two different databases. Each of the databases stores a different type of information for the individuals. The method further includes identifying a set of features common across the at least two different databases. The method further includes generating a unique identification for each of the individuals in the retrieved de-identified records based on the set of features. The method further includes computing a rarity coefficient for each of the individuals based on the set of features. The method further includes matching the de-identified entities across the at least two different databases based on the rarity coefficients. The method further includes matching the de-identified patient records for a set of matched de-identified entities. The method further includes constructing a database with one or more sets of the matched de-identified records.

Abstract: A method includes receiving a first set of de-identified records for individuals from a first type of database for a first set of entities. The first type of database does not include longitudinal information that links the first set of de-identified records across the first set of entities. The method includes receiving a second set of de-identified records for a single individual from a second type of database for a second set of entities. The second type of database includes longitudinal information that links the second set of de-identified records across the second set of entities including over time. The method includes integrating the first type of databases and the second type of databases, which matches the individuals and the single individual. The method includes adding longitudinal information to the first type of database for the individuals based on the longitudinal information of the second type of database.

Abstract: A magnetic resonance spectroscopy assembly includes a magnet to generate a steady magnetic field, an RF transmit/receive antenna to transmit an RF excitation field into an examination region and acquire magnetic resonance signals from the examination region and a magnetic resonance spectrometer coupled to the RF transmit/receive antenna to collect magnetic resonance spectroscopy data from the magnetic resonance signals. An interventional instrument is provided with the assembly. The interventional instruments carries an optical module to generate photonic radiation endowed with orbital optical momentum (OAM).

Abstract: An image guided radiation therapy system comprises a radiation source to generate radiation. Radiation optics forms a therapeutic radiation beam from the therapeutic radiation from the radiation source. An imaging system forms an image of a target zone to control the radiation optics to direct the therapeutic radiation beam onto the target zone. The radiation optics is provided with an optics module configured to generate an imaging photonic beam endowed with optical angular momentum. The imaging system comprises a magnetic resonance assembly to receive magnetic resonance signals the from the target zone generated by imaging photonic beam endowed with optical angular momentum.