# US 12,216,799 B2 · Systems and methods for computing with private healthcare data

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,216,799 B2
- Document ID: US12216799B2
- Status: Granted
- Kind: B2
- Application number: 18/381,873
- Filed: 2023-10-19
- Published: 2024-04-11
- Granted: 2025-02-04
- Earliest claimed: 2019-06-21
- Assignees: nference, Inc.
- Applicants: Nference, Inc.
- Inventors: Sankar Ardhanari; Karthik Murugadoss; Murali Aravamudan; Ajit Rajasekharan
- Web page: https://patentgrep.com/patent/US12216799B2

## CPC

- G06F 21/53: by executing in a restricted environment, e.g. sandbox or secure virtual machine
- G06F 21/6254: {by anonymising data, e.g. decorrelating personal data from the owner's identification}
- G06F 21/602: {Providing cryptographic facilities or services}
- G06F 21/6245: {Protecting personal data, e.g. for financial or medical purposes}
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G16H 10/60: for patient-specific data, e.g. for electronic patient records
- G16H 15/00: ICT specially adapted for medical reports, e.g. generation or transmission thereof
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G16H 70/40: relating to drugs, e.g. their side effects or intended usage
- G16H 70/60: relating to pathologies
- G06N 3/045: Combinations of networks
- G06N 5/02: Knowledge representation; Symbolic representation
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

Techniques are provided for computing with private healthcare data. The techniques include a de-identification method including receiving a text sequence; providing the text sequence to a plurality of entity tagging models, each of the plurality of entity tagging models being trained to tag one or more portions of the text sequence having a corresponding entity type; tagging one or more entities in the text sequence using the plurality of entity tagging models; and obfuscating each entity among the one or more tagged entities by replacing the entity with a surrogate, the surrogate being selected based on one or more attributes of the entity and maintaining characteristics similar to the entity being replaced.

## Claims

### Claim 1

A de-identification method comprising:
receiving a plurality of data sets, wherein the plurality of data sets comprises:
a first data set, wherein the first data set comprises a labeled data set for one or more entity types; and
a second data set, wherein the training data set comprises an unlabeled data set for the one or more entity types;
determining one machine-learning model from a plurality of machine-learning models for each of one or more entity types;
fine-tuning the determined machine-learning model for each of the one or more entity types, wherein fine-tuning the determined machine-learning model comprises:
creating a plurality of training data sets, wherein the plurality of training data sets comprises:
a first training data set, wherein the first training data set comprises the first data set; and
a second training data set, wherein the second training data set comprises the second data set;
training the determined machine-learning model using the first training data set;
validating the trained machine-learning model, wherein validating the trained machine learning model further comprises:
generating a recall score for each entity type of the one or more entity types;
comparing the recall score to a threshold for the recall score for each entity type of the one or more entity types; and
updating the trained machine-learning model using the second training data set as a function of the validation; and
obfuscating the second data set using the fine-tuned machine-learning model.

### Claim 2 (depends on 1)

The de-identification method of claim 1, wherein obfuscating the second data set further comprises replacing two or more entities that refer to a common subject with a common surrogate.

### Claim 3 (depends on 2)

The de-identification method of claim 2, further comprising:
selecting the common surrogate based on one or more attributes of the two or more entities.

### Claim 4 (depends on 2)

The de-identification method of claim 2, further comprising:
selecting the common surrogate based on a gender associated with the two or more entities.

### Claim 5 (depends on 2)

The de-identification method of claim 2, further comprising:
selecting the common surrogate based on an ethnicity associated with the two or more entities.

### Claim 6 (depends on 1)

The de-identification method of claim 1, wherein validating the trained machine learning model further comprises:
generating a precision score for each entity type of the one or more entity types; and
comparing the precision score to a threshold for the precision score for each entity type of the one or more entity types.

### Claim 7 (depends on 1)

The de-identification method of claim 1, wherein validating the trained machine learning model further comprises:
generating a F-score for each entity type of the one or more entity types; and
comparing the F-score to a threshold for the F-score for each entity type of the one or more entity types.

### Claim 8 (depends on 1)

The de-identification method of claim 1, further comprising:
updating the trained machine-learning model as a function of a comparison of an average of a F-score, a precision score and the recall score to a threshold success percentage.

### Claim 9 (depends on 1)

The de-identification method of claim 1, wherein:
the one or more entity types comprises two or more personal names; and
obfuscating the second data set further comprises replacing each of the two or more personal names with a different surrogate.

### Claim 10 (depends on 1)

The de-identification method of claim 1, wherein obfuscating the second data set further comprises replacing two or more entities that refer to a common person with surrogates that match a gender associated with the common person.

### Claim 11 (depends on 1)

The de-identification method of claim 1, wherein obfuscating the second data set further comprises replacing two or more entity types that refer to a common person with surrogates that match an ethnicity associated with the common person.

### Claim 12 (depends on 1)

The de-identification method of claim 1, wherein:
obfuscating the second data set further comprises replacing two or more entities that represent dates with surrogate dates, wherein:
the surrogate dates are based on the two or more entity types altered by a random value.

### Claim 13 (depends on 12)

The de-identification method of claim 12, wherein the surrogate dates associated with a common patient are altered by the same random value.

### Claim 14 (depends on 1)

The de-identification method of claim 1, wherein obfuscating the second data set further comprises scrambling two or more entities that represent numeric identifiers with random values to scramble the numeric identifiers.

### Claim 15 (depends on 1)

The de-identification method of claim 1, wherein the one or more entity types comprises at least a portion of an electronic health record.

### Claim 16 (depends on 1)

The de-identification method of claim 1, further comprising:
receiving a text sequence; and
tagging one or more entities in the text sequence.

### Claim 17 (depends on 16)

The de-identification method of claim 16, further comprising:
aggregating the tagged entities from the text sequence; and
passing the aggregated tagged entities through one or more dreg filters, wherein each of the one or more dreg filters is configured to filter a corresponding entity type based on a rule-based template.

### Claim 18 (depends on 17)

The de-identification method of claim 17, further comprising, creating the rule-based template, wherein creating the rule-based template comprises:
mapping each of one or more portions of the text sequence to a corresponding syntax template;
identifying a candidate syntax template based on a second machine learning model that infers one or more candidate syntax templates based on the one or more portions of the text sequence; and
creating the rule-based template from the candidate syntax template by replacing each of the one or more tagged entities in the portion of the text sequence corresponding to the candidate template with a corresponding syntax token.

### Claim 19 (depends on 17)

The de-identification method of claim 17, wherein each of the one or more dreg filters is further configured to filter the corresponding entity type based on a pattern-based filter.