# US 2021/0343012 A1 · Medical image classification method, model training method, computing device, and storage medium

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

- Number: US 2021/0343012 A1
- Document ID: US20210343012A1
- Status: Published application
- Kind: A1
- Application number: 17/375,177
- Filed: 2021-07-14
- Published: 2021-11-04
- Earliest claimed: 2019-06-21
- Assignees: Tencent Technology (Shenzhen) Company Limited
- Applicants: Tencent Technology (Shenzhen) Company Limited
- Inventors: Kaiwen XIAO; Xiao HAN; Hu YE; Niyun ZHOU
- Web page: https://patentgrep.com/patent/US20210343012A1

## CPC

- G06T 1/20: Processor architectures; Processor configuration, e.g. pipelining
- G06T 7/0012: {Biomedical image inspection}
- G06T 3/4007: based on interpolation, e.g. bilinear interpolation (image demosaicing G06T3/4015; edge-driven or edge-based scaling G06T3/403)
- G06T 7/90: Determination of colour characteristics
- G16H 30/20: for handling medical images, e.g. DICOM, HL7 or PACS
- G16H 30/40: for processing medical images, e.g. editing
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G06T 2207/10056: Microscopic image
- G06T 2207/20076: Probabilistic image processing
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30004: Biomedical image processing
- G06T 2207/30024: Cell structures in vitro; Tissue sections in vitro
- G06T 2207/30096: Tumor; Lesion
- G06T 2207/30168: Image quality inspection
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

This application describes a medical image classification method, a model training method, and a server. The medical image classification method includes: obtaining, by a device, a medical image data set. The device includes a memory storing instructions and a processor in communication with the memory. The method includes performing, by the device, quality analysis on the medical image data set, to extract feature information of a medical image in the medical image data set; and classifying, by the device, the medical image data set based on the feature information and by using a pre-trained deep learning network for performing anomaly detection and classification, to obtain a classification result.

## Claims

### Claim 1

A method for classifying a medical image, the method comprising:
obtaining, by a device comprising a memory storing instructions and a processor in communication with the memory, a medical image data set;
performing, by the device, quality analysis on the medical image data set, to extract feature information of a medical image in the medical image data set; and
classifying, by the device, the medical image data set based on the feature information and by using a pre-trained deep learning network for performing anomaly detection and classification, to obtain a classification result.

### Claim 2 (depends on 1)

The method according to claim 1, wherein:
the quality analysis comprises at least one of the following: hue-saturation-lightness analysis, definition analysis, texture analysis, or entropy value analysis; and
the performing quality analysis on the medical image data set, to extract the feature information of the medical image comprises:
performing, by the device, at least one of the following: the hue-saturation-lightness analysis, the definition analysis, the texture analysis, or the entropy value analysis on the medical image data set, to extract the feature information of the medical image, wherein the feature information comprises at least one of the following: hue feature information, saturation feature information, lightness feature information, a definition index, a grayscale edge, or an entropy value.

### Claim 3 (depends on 2)

The method according to claim 2, wherein the classifying the medical image data set based on the feature information and by using the pre-trained deep learning network for performing anomaly detection and classification comprises:
removing, by the device, an irrelevant image from the medical image data set according to the feature information, wherein the irrelevant image comprises a non-medical image; and
classifying, by the device using the pre-trained deep learning network for performing anomaly detection and classification, the medical image data set from which the irrelevant image is removed.

### Claim 4 (depends on 2)

The method according to claim 2, wherein:
performing the hue-saturation-lightness analysis comprises converting values of a red coordinate, a green coordinate, and a blue coordinate of the medical image from an RGB space to an HSV space, to obtain hue feature information, saturation feature information, and lightness feature information;
performing the definition analysis comprises:
calculating a value of a digital image matrix of the medical image,
performing convolution on the value of the digital image matrix and a 5×5 Gaussian convolution kernel to obtain a convolution value, and
calculating a definition index of the medical image based on a minimum mean square error of the value of the digital image matrix and the convolution value;
performing the texture analysis comprises extracting a grayscale edge of the medical image by using a Sobel verification operator; and
performing the entropy value analysis comprises calculating an entropy value according to a length of the medical image, a width of the medical image, and a quantity of times and a probability that a grayscale value of a center pixel in a sliding window and an average of grayscale values other than that of the center pixel in the sliding window occur in the medical image.

### Claim 5 (depends on 1)

The method according to claim 1, further comprising:
before performing the quality analysis:
performing, by the device, file check on the medical image data set to determine whether the medical image in the medical image data set is capable of being parsed; and
in response to determining that whether the medical image in the medical image data set is capable of being parsed, decoding, by the device, the medical image data set to convert the medical image into a digital image matrix.

### Claim 6 (depends on 1)

The method according to claim 1, further comprising:
before classifying the medical image data set, performing, by the device, interpolation and normalization on the medical image in the medical image data set, wherein the interpolation comprises bilinear interpolation.

### Claim 7 (depends on 1)

The method according to claim 1, further comprising:
training, by the device, the pre-trained deep learning network by selecting an Inception V3 model.

### Claim 8

An apparatus for classifying a medical image, the apparatus comprising:
a memory storing instructions; and
a processor in communication with the memory, wherein, when the processor executes the instructions, the processor is configured to cause the apparatus to:
obtain a medical image data set,
perform quality analysis on the medical image data set, to extract feature information of a medical image in the medical image data set, and
classify the medical image data set based on the feature information and by using a pre-trained deep learning network for performing anomaly detection and classification, to obtain a classification result.

### Claim 9 (depends on 8)

The apparatus according to claim 8, wherein:
the quality analysis comprises at least one of the following: hue-saturation-lightness analysis, definition analysis, texture analysis, or entropy value analysis; and
when the processor is configured to cause the apparatus to perform quality analysis on the medical image data set, to extract the feature information of the medical image, the processor is configured to cause the apparatus to:
perform at least one of the following: the hue-saturation-lightness analysis, the definition analysis, the texture analysis, or the entropy value analysis on the medical image data set, to extract the feature information of the medical image, wherein the feature information comprises at least one of the following: hue feature information, saturation feature information, lightness feature information, a definition index, a grayscale edge, or an entropy value.

### Claim 10 (depends on 9)

The apparatus according to claim 9, wherein, when the processor is configured to cause the apparatus to classify the medical image data set based on the feature information and by using the pre-trained deep learning network for performing anomaly detection and classification, the processor is configured to cause the apparatus to:
remove an irrelevant image from the medical image data set according to the feature information, wherein the irrelevant image comprises a non-medical image; and
classify, by using the pre-trained deep learning network for performing anomaly detection and classification, the medical image data set from which the irrelevant image is removed.

### Claim 11 (depends on 9)

The apparatus according to claim 9, wherein:
when the processor is configured to cause the apparatus to perform the hue-saturation-lightness analysis, the processor is configured to cause the apparatus to convert values of a red coordinate, a green coordinate, and a blue coordinate of the medical image from an RGB space to an HSV space, to obtain hue feature information, saturation feature information, and lightness feature information;
when the processor is configured to cause the apparatus to perform the definition analysis, the processor is configured to cause the apparatus to:
calculate a value of a digital image matrix of the medical image,
perform convolution on the value of the digital image matrix and a 5×5 Gaussian convolution kernel to obtain a convolution value, and
calculate a definition index of the medical image based on a minimum mean square error of the value of the digital image matrix and the convolution value;
when the processor is configured to cause the apparatus to perform the texture analysis, the processor is configured to cause the apparatus to extract a grayscale edge of the medical image by using a Sobel verification operator; and
when the processor is configured to cause the apparatus to perform the entropy value analysis, the processor is configured to cause the apparatus to calculate an entropy value according to a length of the medical image, a width of the medical image, and a quantity of times and a probability that a grayscale value of a center pixel in a sliding window and an average of grayscale values other than that of the center pixel in the sliding window occur in the medical image.

### Claim 12 (depends on 8)

The apparatus according to claim 8, wherein, before the processor is configured to cause the apparatus to perform the quality analysis, the processor is configured to further cause the apparatus to:
perform file check on the medical image data set to determine whether the medical image in the medical image data set is capable of being parsed; and
in response to determining that whether the medical image in the medical image data set is capable of being parsed, decode the medical image data set to convert the medical image into a digital image matrix.

### Claim 13 (depends on 8)

The apparatus according to claim 8, wherein, before the processor is configured to cause the apparatus to classify the medical image data set, the processor is configured to further cause the apparatus to:
perform interpolation and normalization on the medical image in the medical image data set, wherein the interpolation comprises bilinear interpolation.

### Claim 14 (depends on 8)

The apparatus according to claim 8, wherein, when the processor executes the instructions, the processor is configured to further cause the apparatus to:
train the pre-trained deep learning network by selecting an Inception V3 model.

### Claim 15

A non-transitory computer-readable storage medium, storing computer-readable instructions, wherein, the computer-readable instructions, when executed by a processor, are configured to cause the processor to perform:
obtaining a medical image data set;
performing quality analysis on the medical image data set, to extract feature information of a medical image in the medical image data set; and
classifying the medical image data set based on the feature information and by using a pre-trained deep learning network for performing anomaly detection and classification, to obtain a classification result.

### Claim 16 (depends on 15)

The non-transitory computer-readable storage medium according to claim 15, wherein:
the quality analysis comprises at least one of the following: hue-saturation-lightness analysis, definition analysis, texture analysis, or entropy value analysis; and
when the computer-readable instructions are configured to cause the processor to perform performing quality analysis on the medical image data set, to extract the feature information of the medical image, the computer-readable instructions are configured to cause the processor to perform:
performing at least one of the following: the hue-saturation-lightness analysis, the definition analysis, the texture analysis, or the entropy value analysis on the medical image data set, to extract the feature information of the medical image, wherein the feature information comprises at least one of the following: hue feature information, saturation feature information, lightness feature information, a definition index, a grayscale edge, or an entropy value.

### Claim 17 (depends on 16)

The non-transitory computer-readable storage medium according to claim 16, wherein, when the computer-readable instructions are configured to cause the processor to perform classifying the medical image data set based on the feature information and by using the pre-trained deep learning network for performing anomaly detection and classification, the computer-readable instructions are configured to cause the processor to perform:
removing an irrelevant image from the medical image data set according to the feature information, wherein the irrelevant image comprises a non-medical image; and
classifying, by using the pre-trained deep learning network for performing anomaly detection and classification, the medical image data set from which the irrelevant image is removed.

### Claim 18 (depends on 16)

The non-transitory computer-readable storage medium according to claim 16, wherein:
when the computer-readable instructions are configured to cause the processor to perform performing the hue-saturation-lightness analysis, the computer-readable instructions are configured to cause the processor to perform converting values of a red coordinate, a green coordinate, and a blue coordinate of the medical image from an RGB space to an HSV space, to obtain hue feature information, saturation feature information, and lightness feature information;
when the computer-readable instructions are configured to cause the processor to perform performing the definition analysis, the computer-readable instructions are configured to cause the processor to perform:
calculating a value of a digital image matrix of the medical image,
performing convolution on the value of the digital image matrix and a 5×5 Gaussian convolution kernel to obtain a convolution value, and
calculating a definition index of the medical image based on a minimum mean square error of the value of the digital image matrix and the convolution value;
when the computer-readable instructions are configured to cause the processor to perform performing the texture analysis, the computer-readable instructions are configured to cause the processor to perform extracting a grayscale edge of the medical image by using a Sobel verification operator; and
when the computer-readable instructions are configured to cause the processor to perform performing the entropy value analysis, the computer-readable instructions are configured to cause the processor to perform calculating an entropy value according to a length of the medical image, a width of the medical image, and a quantity of times and a probability that a grayscale value of a center pixel in a sliding window and an average of grayscale values other than that of the center pixel in the sliding window occur in the medical image.

### Claim 19 (depends on 15)

The non-transitory computer-readable storage medium according to claim 15, wherein, before the computer-readable instructions are configured to cause the processor to perform performing the quality analysis, the computer-readable instructions are configured to further cause the processor to perform:
performing file check on the medical image data set to determine whether the medical image in the medical image data set is capable of being parsed; and
in response to determining that whether the medical image in the medical image data set is capable of being parsed, decoding the medical image data set to convert the medical image into a digital image matrix.

### Claim 20 (depends on 15)

The non-transitory computer-readable storage medium according to claim 15, wherein, before the computer-readable instructions are configured to cause the processor to perform classifying the medical image data set, the computer-readable instructions are configured to further cause the processor to perform:
performing interpolation and normalization on the medical image in the medical image data set, wherein the interpolation comprises bilinear interpolation.
