# US 2026/0038699 A1 · Medical database searching system

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

- Number: US 2026/0038699 A1
- Document ID: US20260038699A1
- Status: Published application
- Kind: A1
- Application number: 19/288,074
- Filed: 2025-08-01
- Published: 2026-02-05
- Earliest claimed: 2024-08-02
- Assignees: Koninklijke Philips N.V.
- Applicants: Koninklijke Philips N.V.
- Inventors: kUMAR SHRESHTHA; LUIZ AUGUSTO FACURY DE SOUZA; MOHAMMED AMINE KHELDOUNI; REMI CHRISTOPHE MANUEL DUPONT; BENOIT REGIS JEAN MARIE CHOFFIN; NICOLA PEZZOTTI
- Web page: https://patentgrep.com/patent/US20260038699A1

## CPC

- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G16H 10/60: for patient-specific data, e.g. for electronic patient records
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G06N 3/08: Learning methods
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The present disclosure provides concepts for searching a medical database comprising electrocardiogram (ECG) data. The method includes obtaining a database comprising a plurality of subject entries, each corresponding to a historic subject and comprising recorded metadata describing characteristics of the respective subject and a recorded vector embedding representing ECG data of the respective subject. A query vector embedding representing ECG data of a query subject is generated with an ECG encoder. The recorded metadata of the plurality of subject entries is compared with query metadata describing characteristics of the query subject, and the recorded vector embeddings are compared with the query vector embedding. One or more similar subject entries are identified based on a result of the comparison. Accordingly, the invention provides a searching means that is designed to take into account characteristics of subjects to identify subjects with similar cardiological conditions, thereby reducing a burden on a clinician whilst reducing a rate of misdiagnosis. For example, this may be particularly useful for triaging subjects.

## Claims

### Claim 1

A method for searching a medical database comprising electrocardiogram, ECG data, the method comprising:
obtaining a database comprising a plurality of subject entries each corresponding to a historic subject, each subject entry comprising recorded metadata describing characteristics of the respective subject and a recorded vector embedding representing ECG data of the respective subject;
generating, with an ECG encoder, a query vector embedding representing ECG data of a query subject;
comparing the recorded metadata of the plurality of subject entries with query metadata describing characteristics of the query subject;
characterized in that the method further comprises:
comparing the recorded vector embeddings with the query vector embedding; and
identifying one or more similar subject entries based on a result of the comparison.

### Claim 2 (depends on 1)

The method of claim 1, wherein comparing the metadata and the vector embeddings, comprises:
filtering the historic subject entries to identify candidate subject entries, the candidate subject entries comprising similar metadata to the metadata of the query subject; and
comparing the vector embeddings of the identified candidate subject entries and the query vector embedding.

### Claim 3 (depends on 2)

The method of claim 2, wherein comparing the vector embeddings comprises processing, with a hierarchical navigable small world, HNSW, algorithm, each of the vector embeddings of the identified candidate subject entries and the query vector embedding.

### Claim 4 (depends on 2)

The method of claim 2, wherein comparing the vector embeddings comprises generating, for each vector embeddings of the identified candidate subject entries, a similarity score indicating a degree of similarity between the query vector embedding and the respective vector embedding of the identified candidate subject entry, and wherein identifying the one or more similar subject entries is based on the generated similarity scores.

### Claim 5 (depends on 2)

The method of claim 2, wherein filtering the historic subject entries is based on a predefined set of criteria defining matching characteristics between the metadata of the subject entry and the metadata of the query subject.

### Claim 6 (depends on 5)

The method of claim 5, wherein the predefined set of criteria is based on user preferences indicating target matching characteristics.

### Claim 7 (depends on 1)

The method of claim 1, wherein the ECG encoder is a neural network adapted to detect features of ECG data and convert the detected features into a vector.

### Claim 8 (depends on 7)

The method of claim 7, wherein the neural network is a vision transformer.

### Claim 9 (depends on 1)

The method of claim 1, further comprising providing at least part of the recorded metadata, the ECG data, reported clinical findings and/or subject outcomes associated with the identified similar subject entries.

### Claim 10 (depends on 1)

The method of claim 1, further comprising ranking the identified similar subject entries based on a degree of similarity between the respective recorded vector embeddings of the similar subject entries and the query vector embedding, and/or the respective recorded metadata of the similar subject entries and the query metadata.

### Claim 11 (depends on 1)

The method of claim 1, wherein the metadata comprises subject demographic information, subject medical information, subject condition information, and subject context information.

### Claim 12 (depends on 1)

The method of claim 1, wherein the vector embeddings are a projection of the ECG data in high dimension space that encodes semantic meaning of the ECG data.

### Claim 13 (depends on 1)

A computer program comprising computer program code means adapted, when said computer program is run on a computer, to implement the method of claim 1.

### Claim 14

A system for searching a medical database, the medical database comprising a plurality of subject entries each corresponding to a historic subject, each subject entry comprising recorded metadata describing characteristics of the respective subject and a recorded vector embedding representing ECG data of the respective subject, the system comprising:
an ECG encoder configured to generate a query vector embedding representing ECG data of a query subject;
a database interface configured to:
compare the recorded metadata of the plurality of subject entries with query metadata describing characteristics of the query subject;
characterized in that the database interface is further configured to:
compare the recorded vector embeddings with the query vector embedding; and
identify one or more similar subject entries based on a result of the comparison.

### Claim 15 (depends on 14)

The system of claim 14, wherein the ECG encoder is a neural network adapted to detect features of ECG data and convert the detected features into a vector.
