CN-121999285-A - Nasal polyp pathological section image identification method, device, electronic equipment and storage medium

Abstract

The disclosure provides a nasal polyp pathological section image identification method, a nasal polyp pathological section image identification device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence and data processing. The nasal polyp pathological section image processing method comprises the steps of performing domain generalization on a nasal polyp pathological section image under a first visual field of a first detection object by adopting a domain generalization model to obtain a nasal polyp pathological section invariant feature map, performing cell category classification and tissue region segmentation on the nasal polyp pathological section invariant feature map by adopting a cell classification model and a tissue region segmentation model to obtain cell types of cells in the nasal polyp pathological section invariant feature map, and performing statistics on an epithelial region, a vascular region and a gland region, and determining the nasal polyp inflammation type of the first detection object according to a statistical result. Thus, the generalization and the accuracy of the identification of the inflammation type of the nasal polyp are improved.

Inventors

• YANG QINTAI
• ZHOU WENHAO
• ZHANG HE
• WU TONG
• HUANG XUEKUN
• ZHANG YANA
• SHI ZHAOHUI
• YANG LIN
• CUI LEI
• LUO XIN
• Hua Zixuan
• CHEN JIANNING
• LIU ZIFENG
• YUAN TIAN
• KANG NING
• Guo Zezhi
• LIANG GUIXIAN

Assignees

• 中山大学附属第三医院

Dates

Publication Date

20260508

Application Date

20260122

Claims (10)

1. 1. The nasal polyp pathological section image identification method is characterized by comprising the following steps of: Performing domain generalization on the nasal polyp pathological section image in the first view of the first detection object by adopting a domain generalization model to obtain a nasal polyp pathological domain invariant feature map; Classifying cell types of the nasal polyp pathological domain invariant feature map by adopting a cell classification model to obtain cell types of cells in the nasal polyp pathological domain invariant feature map; Performing tissue region segmentation on the nasal polyp pathological region invariant feature map by adopting a tissue region segmentation model to obtain an epithelial region, a blood vessel region and a gland region in the nasal polyp pathological region invariant feature map; counting the number of cells of each cell type in the first field of view based on the cell type of each cell in the nasal polyp pathological domain invariant feature map; Based on an epithelial region, a vascular region and a gland region in the nasal polyp pathological region invariant feature map, calculating an epithelial region area, a vascular region area and a gland region area under the first visual field; Determining the nasal polyp inflammation type of the first detection object based on the cell number of each cell type under the first visual field, and the epithelial region area, the vascular region area and the glandular region area.
2. 2. The method of claim 1, wherein the domain generalization model comprises a first encoder consisting of a backbone network, and a first decoder, the method comprising: Inputting a nasal polyp pathological section image sample under a second visual field of a second detection object into the backbone network to obtain a low-order characteristic provided by a second network and a high-order characteristic provided by a fifth network in the backbone network, wherein the backbone network comprises 5 networks which are input to output and gradually decrease in network size; Carrying out random convolution on the low-order features to obtain disturbance features; Extracting the characteristics of the disturbance characteristics through a leakage mixing network to obtain enhanced characteristics; Connecting the high-order features with the enhancement features, and inputting the features obtained after connection into the first decoder to obtain a prediction domain invariant feature map output by the first decoder; And based on the difference between the prediction domain invariant feature map and the real domain invariant feature map corresponding to the nasal polyp pathological section image sample, carrying out network parameter adjustment on the backbone network, the first decoder and the leakage mixed network.
3. 3. The method as recited in claim 2, further comprising: Performing domain classification on the high-order features by using a classifier to obtain a prediction domain source of the nasal polyp pathological section image sample; and adjusting network parameters of the classifier and the first encoder based on the difference between the predicted domain source and the corresponding real domain source of the nasal polyp pathological section image sample.
4. 4. The method according to claim 2 or 3, wherein the performing, using a domain generalization model, domain generalizing the nasal polyp pathological section image in the first field of view of the first detection object to obtain a nasal polyp pathological domain invariant feature map comprises: performing pixel clustering on the nasal polyp pathological section image under the first view of the detection object to obtain a foreground cell pixel area and a background fragment pixel area; dividing the nasal polyp pathological section image into a plurality of rectangular images; extracting rectangular images intersecting the foreground cell pixel areas from the plurality of rectangular images, and forming the extracted rectangular images into foreground cell image areas; And performing domain generalization on the foreground cell image area by adopting a domain generalization model to obtain the nasal polyp pathological domain invariant feature map.
5. 5. The method of claim 4, wherein the performing the domain generalization on the foreground cell image region using a domain generalization model to obtain the nasal polyp pathology domain invariant feature map comprises: inputting the foreground cell image area into a first encoder in the domain generalization model to obtain high-order features output by the first encoder; And inputting the high-order features into a first decoder in the domain generalization model to obtain the nasal polyp pathological domain invariant feature map output by the first decoder.
6. 6. The method of claim 1, wherein the cell classification model comprises a feature extraction module, an attention module, a linear regression module, a linear classification module, and a one-to-one matching module, wherein the feature extraction module comprises a feature pyramid network, wherein the employing the cell classification model to classify the cell class of the nasal polyp pathology domain invariant feature map, obtaining the cell type of each cell in the nasal polyp pathology domain invariant feature map comprises: inputting the nasal polyp pathological domain invariant feature map into the feature extraction module to obtain each tower layer feature output by a feature pyramid network in the feature extraction module; inputting each tower layer characteristic into the attention module to obtain an attention aggregation characteristic output by the attention module; Inputting the attention aggregation characteristics into the linear regression module to obtain linear regression characteristics of each cell output by the linear regression module; inputting the attention aggregation characteristics into the linear classification module to obtain linear classification characteristics of each cell output by the linear classification module; And respectively convoluting the linear regression feature and the linear classification feature of each cell to obtain the convolution feature of each cell, and carrying out one-to-one matching classification on the convolution feature of each cell through the one-to-one matching module to obtain the cell type of each cell.
7. 7. The method of claim 1, wherein the tissue region segmentation model employs a DeepLabv3+ model.
8. 8. A nasal polyp pathological section image recognition device, characterized by comprising: the domain generalization module is used for performing domain generalization on the nasal polyp pathological section image under the first visual field of the first detection object by adopting a domain generalization model to obtain a nasal polyp pathological domain invariant feature map; The cell classification module is used for classifying cell types of the nasal polyp pathological domain invariant feature map by adopting a cell classification model to obtain cell types of cells in the nasal polyp pathological domain invariant feature map; the region segmentation module is used for carrying out tissue region segmentation on the nasal polyp pathological region invariant feature map by adopting a tissue region segmentation model to obtain an epithelial region, a blood vessel region and a gland region in the nasal polyp pathological region invariant feature map; a first statistics module, configured to count, based on cell types of each cell in the nasal polyp pathology domain invariant feature map, a number of cells of each cell type in the first field of view; the second statistics module is used for counting the area of the epithelium region, the area of the blood vessel region and the area of the gland region under the first visual field based on the epithelium region, the blood vessel region and the gland region in the nasal polyp pathology domain invariant feature map; and the inflammation type determining module is used for determining the nasal polyp inflammation type of the first detection object based on the cell number of each cell type in the first visual field, and the area of an epithelial area, the area of a blood vessel area and the area of a gland area.
9. 9. An electronic device, comprising: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.
10. 10. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-7.

Description

Nasal polyp pathological section image identification method, device, electronic equipment and storage medium Technical Field The present disclosure relates to the field of artificial intelligence and image processing technology. The disclosure relates specifically to a nasal polyp pathological section image identification method, a nasal polyp pathological section image identification device, an electronic device and a storage medium. Background Nasal polyps (nasal polyp) are common diseases of the nose, which are formed by gradual sagging of the nasal sinus mucosa under the force of gravity, which is extremely drooped. Nasal polyps are detected mainly by three modes, (1) intranasal endoscopy, which is to observe viscous or mucopurulent secretions from middle nasal passages and olfactory clefts, nasal mucosa congestion, edema or polyps. Typically this will only be a primary screening mode. (2) Imaging examination the CT scan of the sinuses can show the sinus ostial and nasal meatus complex and/or the inflammatory lesions of the sinuses mucosa. (3) Pathological examination, which is a gold standard for various examinations, can specifically judge the pathological type of nasal polyp. With the progress of artificial intelligence and advanced image processing technology, an artificial intelligence auxiliary diagnosis technology based on digital images is innovatively and integrally developed in different professional fields of medicine, and the artificial intelligence technology has gradually become a powerful auxiliary tool for accurate diagnosis in the medical field. Digital pathology is an important component in artificial intelligence aided diagnosis, namely, a physical pathological tissue slide is converted into a digital pathological slide by using a digital pathological scanning instrument, and is checked and analyzed on a display by software. Common digital pathology image analysis tasks include four general categories, 1) slice classification, such as endometrial cancer molecular typing, 2) target detection, such as thin-layer liquid-based cytology (TCT) lesion cell detection, 3) region segmentation, such as gland segmentation, 4) cell detection, such as tumor cell PD-L1 scoring, and the like. However, existing artificial intelligence nasal polyp identification methods are mainly used for identifying pathological subtypes, lack accurate identification on a cell level, and lack the ability to identify dense cell areas due to algorithm limitations. Disclosure of Invention The present disclosure provides a nasal polyp pathological section image recognition method, apparatus, electronic device and storage medium. According to an aspect of the present disclosure, there is provided a nasal polyp pathological section image recognition method, including: Performing domain generalization on the nasal polyp pathological section image in the first view of the first detection object by adopting a domain generalization model to obtain a nasal polyp pathological domain invariant feature map; Classifying cell types of the nasal polyp pathological domain invariant feature map by adopting a cell classification model to obtain cell types of cells in the nasal polyp pathological domain invariant feature map; Performing tissue region segmentation on the nasal polyp pathological region invariant feature map by adopting a tissue region segmentation model to obtain an epithelial region, a blood vessel region and a gland region in the nasal polyp pathological region invariant feature map; counting the number of cells of each cell type in the first field of view based on the cell type of each cell in the nasal polyp pathological domain invariant feature map; Based on an epithelial region, a vascular region and a gland region in the nasal polyp pathological region invariant feature map, calculating an epithelial region area, a vascular region area and a gland region area under the first visual field; Determining the nasal polyp inflammation type of the first detection object based on the cell number of each cell type under the first visual field, and the epithelial region area, the vascular region area and the glandular region area. According to another aspect of the present disclosure, there is provided a nasal polyp pathological section image recognition apparatus, including: the domain generalization module is used for performing domain generalization on the nasal polyp pathological section image under the first visual field of the first detection object by adopting a domain generalization model to obtain a nasal polyp pathological domain invariant feature map; The cell classification module is used for classifying cell types of the nasal polyp pathological domain invariant feature map by adopting a cell classification model to obtain cell types of cells in the nasal polyp pathological domain invariant feature map; the region segmentation module is used for carrying out tissue region segmentation on the nasal polyp patholo