# US 12,183,105 B2 · Method to improve instance selection in bootstrapping framework for concept extraction from text documents

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

- Number: US 12,183,105 B2
- Document ID: US12183105B2
- Status: Granted
- Kind: B2
- Application number: 17/781,074
- Filed: 2020-12-18
- Published: 2023-01-19
- Granted: 2024-12-31
- Earliest claimed: 2019-12-18
- Assignees: Koninklijke Philips N.V.
- Applicants: Koninklijke Philips N.V.
- Inventors: Ashequl Qadir; Kathy Mi Young Lee; Claire Yunzhu Zhao; Aaditya Prakash; Minnan Xu
- Web page: https://patentgrep.com/patent/US12183105B2

## CPC

- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06V 30/413: Classification of content, e.g. text, photographs or tables
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G16H 15/00: ICT specially adapted for medical reports, e.g. generation or transmission thereof
- G16H 30/40: for processing medical images, e.g. editing
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

A system and method for unsupervised training of a text report identification machine learning model, including: labeling a first set of unlabeled text reports using a seed dictionary to identify concepts in the unlabeled text reports; inputting images associated with the first set of seed-labeled text reports into an auto-encoder to obtain an encoded first set of images; calculating a set of first correlation matrices as a dot product of the first encoded images with their associated text report features; determining a distance between the set of first correlation matrices and a filter bank value associated with the auto-encoder; identifying a first set of validated images as the images in the first set of images that have a distance less than a threshold value; and training the text report machine learning model using the labeled text reports associated with the set of first validated images.

## Claims

### Claim 1

A method for weakly-supervised training of a text report identification machine learning model, comprising:
labeling a first set of unlabeled text reports based on matching concept phrases of a seed dictionary with concepts in the unlabeled text reports;
inputting images associated with the first set of labeled text reports into an auto-encoder to obtain an encoded first set of images;
calculating a set of first correlation matrices as a dot product of the first encoded images with their associated text report features;
determining a distance between the set of first correlation matrices and a filter bank value associated with the auto-encoder;
identifying a first set of validated images as the images in the first set of images that have a distance less than a threshold value; and
training the text report machine learning model using the labeled text reports associated with the set of first validated images.

### Claim 2 (depends on 1)

The method of claim 1, further comprising
labeling a second set of unlabeled text reports using trained text report machine learning model;
inputting images associated with the second set labeled text reports into an auto-encoder to obtain an encoded second set of images;
calculating a set of second correlation matrices as a dot product of the second encoded images with their associated text report features;
determining a distance between the set of second correlation matrices and the filter bank value associated with the auto-encoder;
identifying a second set of validated images as the images in the second set of images that have a distance less than a threshold value; and
training the text report machine learning model using the labeled text reports associated with the first set of validated images and the second set of validated images.

### Claim 3 (depends on 2)

The method of claim 2, further comprising repeating the steps of claim 2 with additional sets of unlabeled data.

### Claim 4 (depends on 3)

The method of claim 3, further comprising completing training text identification machine learning model when no further unlabeled data is available.

### Claim 5 (depends on 3)

The method of claim 3, further comprising determining performance of the text identification model using a validation set of data for each iteration of claim 3.

### Claim 6 (depends on 4)

The method of claim 4, further comprising completing training text identification machine learning model when performance between iterations is less than a performance change threshold value.

### Claim 7 (depends on 1)

The method of claim 1, further comprising training the auto-encoder using a loss function based upon error between an input and output image of the auto-encoder and correlation matrix.

### Claim 8 (depends on 7)

The method of claim 7, wherein the loss function is calculated as:
J=MSE−λI′T·T,
where J is the loss function, MSE is mean square error between the input image and the output image of the auto-encoder, λ is a hyperparameter, I′T·T is a correlation matrix, I′T is a transpose of an encoded image, and T is the text report features.

### Claim 9

A system for training a text report identification machine learning model and an image identification machine learning model, comprising:
a memory; and
a processor connected to the memory, the processor configured to:
label a first set of unlabeled text reports based on matching concept phrases of a seed dictionary with concepts in the unlabeled text reports;
input images associated with the first set of labeled text reports into an auto-encoder to obtain an encoded first set of images;
calculate a set of first correlation matrices as a dot product of the first encoded images with their associated text report features;
determine a distance between the set of first correlation matrices and a filter bank value associated with the auto-encoder;
identify a first set of validated images as the images in the first set of images that have a distance less than a threshold value; and
train the text report machine learning model using the labeled text reports associated with the set of first validated images.

### Claim 10 (depends on 9)

The system of claim 9, wherein the processor is further configured to:
label a second set of unlabeled text reports using trained text report machine learning model;
input images associated with the second set of labeled text reports into an auto-encoder to obtain an encoded second set of images;
calculate a set of second correlation matrices as a dot product of the second encoded images with their associated text report features;
determine a distance between the set of second correlation matrices and the filter bank value associated with the auto-encoder;
identify a second set of validated images as the images in the second set of images that have a distance less than a threshold value; and
train the text report machine learning model using the labeled text reports associated with the first set of validated images and the second set of validated images.

### Claim 11 (depends on 10)

The system of claim 10, wherein the processor is further configured to: repeat the steps of claim 10 with additional sets of unlabeled data.

### Claim 12 (depends on 11)

The system of claim 11, wherein the processor is further configured to complete training text identification machine learning model when no further unlabeled data is available.

### Claim 13 (depends on 11)

The system of claim 11, wherein the processor is further configured to determine performance of the text identification model using a validation set of data for each iteration of claim 11.

### Claim 14 (depends on 13)

The system of claim 13, wherein the processor is further configured to complete training text identification machine learning model when performance change between iterations is less than a performance threshold value.

### Claim 15 (depends on 9)

The system of claim 9, wherein the processor is further configured to train the auto-encoder using a loss function based upon error between an input and output image of the auto-encoder and a correlation matrix.

### Claim 16 (depends on 15)

The system of claim 15, wherein the loss function is calculated as:
J=MSE−λI{circumflex over ( )}′T·T,
where J is the loss function, MSE is mean square error between the input image and the output image of the auto-encoder, λ is a hyperparameter, I′T·T is a correlation matrix, I′T is a transpose of an encoded image, and T is the text report features.