Pathological section analyzer with large field of view, high throughput and high resolution

Patent metadata

Field	Value
Publication number	US-12548115-B2
Application number	US-202318104812-A
Country	US
Kind code	B2
Filing date	Feb 2, 2023
Priority date	Dec 7, 2020
Publication date	Feb 10, 2026
Grant date	Feb 10, 2026

Abstract

A large-field-of-view, high-throughput and high-resolution pathological section analyzer includes an image collector for collecting a set of computing microscopic images of a pathological section sample; a data preprocessing circuit for iteratively updating the set of computing microscopic images by a multi-height phase recovery algorithm to obtain a low-resolution reconstructed image; an image super-resolution circuit for super-resolving the low-resolution reconstructed image according to a pre-trained super-resolution model to obtain a high-resolution reconstructed image; and an image analysis circuit for automatically analyzing the high-resolution reconstructed image according to different tasks, and specifically selecting different analysis models according to the different tasks to obtain corresponding auxiliary diagnosis results. Imaging visual field of the pathological section analyzer is hundreds of times that of the traditional optical microscope, a deep learning network is adopted to analyze pathological conditions of unstained pathological sections, so that the analysis process of pathological sections is simplified.

First claim

What is claimed is:

1 . A pathological section analyzer, comprising: an image collector, a data preprocessing circuit, an image super-resolution circuit and an image analysis circuit;
wherein a use process of the pathological section analyzer comprises: step 1, obtaining an unstained pathological section sample from clinic; step 2, collecting computing microscopic images of the pathological section sample in step 1 at different heights by using the image collector; step 3, inputting the computing microscopic images at the different heights in step 2 into the data preprocessing circuit to obtain a low-resolution reconstructed image after twin images are removed; step 4, inputting the low-resolution reconstructed image in step 3 into the image super-resolution circuit to obtain a high-resolution reconstructed image; and step 5, inputting the high-resolution reconstructed image in step 4 into the image analysis circuit to obtain an auxiliary diagnosis result of the pathological section sample in step 1; wherein the image super-resolution circuit is configured to perform the following steps: step 41, cropping the low-resolution reconstructed image into image patches with a size of 512*512 pixels in a fixed order, with lengths and widths overlapping by ⅔ pixels; step 42, inputting the image patches in step 41 into a pre-trained super-resolution model to obtain high-resolution image patches; step 43, replacing pixel values of ⅓ areas at boundaries of the left and right adjacent high-resolution image patches with pixel values of a ⅓ length or width area at a middle of the high-resolution image patch to eliminate boundary blur caused by truncation effect; step 44, recovering the high-resolution image patches without the boundary blur obtained in step 43 into the high-resolution reconstructed image according to the fixed order in step 41, and thereby realizing super-resolution of the low-resolution reconstructed image; wherein the high-resolution reconstruction image is also referred to as a whole slide image (WSI) in clinical electronic scanning.

2 . The pathological section analyzer according to claim 1 , wherein the image collector comprises: a lensfree coaxial computing microscopic multi-height image collection system, a one-dimensional translation stage and a two-dimensional electric translation stage;
wherein the lensfree coaxial computing microscopic multi-height image collection system comprises a monochromatic laser light source, a pinhole, the pathological section sample, an objective table, a complementary metal oxide semiconductor (CMOS) image sensor and an optical connection component; wherein a spherical wave emitted by the monochromatic laser light source is filtered by the pinhole and then transmitted to a plane of the pathological section sample placed on the objective table to generate a surface wave and a reference wave, and the surface wave and the reference wave interfere to form a hologram, which is recorded by the CMOS image sensor to obtain a computing microscopic image; wherein the one-dimensional translation stage and the two-dimensional electric translation stage are an integrated optical component, which are respectively configured to adjust the CMOS image sensor in the lensfree coaxial computing microscopic multi-height image collection system back and forth along an optical path direction, and adjust a position of the monochromatic laser light source, thereby to make a light wave emitted by the monochromatic laser light source illuminate the pathological section sample; wherein the lensfree coaxial computational microscope multi-height image collection system is configured to collect the computing microscopic images at the different heights, and specifically configured to perform the following steps: step 21, adjusting the two-dimensional electric translation stage controlling the monochromatic laser light source and the one-dimensional translation stage controlling the CMOS image sensor to make the light wave emitted by the monochromatic laser light source completely cover a photosensitive plane of the CMOS image sensor; step 22, recording a computing microscopic image captured by the CMOS image sensor when the pathological section sample is not placed on the objective table; step 23, placing the pathological section sample on the objective table, rotating a knob of the one-dimensional translation stage forward at a fixed interval along a light wave transmission direction, and capturing the computing microscopic images at the different heights by the CMOS image sensor; and step 24, taking the computing microscopic images collected in steps 22 and 23 as a set of computing microscopic images corresponding to the pathological section sample.

3 . The pathological section analyzer according to claim 2 , wherein the data preprocessing circuit is configured to use a multi-height phase recovery algorithm to iteratively update the set of computing microscopic images of the pathological section sample to eliminate interference of the twin images, and specifically configured to perform the following steps:
step 31, registering the computing microscopic images at the different heights to eliminate lateral drift among the computing microscopic images at the different heights when rotating the knob of the one-dimensional translation stage; step 32, iteratively updating the registered computing microscopic images at the different heights according to the multi-height phase recovery algorithm to obtain the low-resolution reconstructed image after the twin images are removed.

4 . The pathological section analyzer according to claim 3 , wherein the pre-trained super-resolution model is obtained by using different training methods according to different sources of actually available hematoxylin-eosin (HE) staining WSIs; when HE-staining WSIs corresponding to tissue areas adjacent to the pathological section sample are obtained, the pre-trained super-resolution model is obtained by training with a training strategy based on registration mapping, or when any HE staining WSI irrelevant to the pathological section sample is obtained, the pre-trained super-resolution model is obtained by training with a training strategy based on modeling mapping;
wherein the training strategy based on registration mapping is obtained by the following steps: step 61, obtaining low-resolution reconstructed images of pathological section samples; step 62, searching and intercepting a HE staining WSI which is consistent with an area of a corresponding one of the low-resolution reconstructed images in step 61 from the adjacent HE staining WSIs of each of the pathological section samples; step 63, interpolating and scaling each the low-resolution reconstructed image in step 61 with a size of the corresponding HE staining WSI in step 62 as a reference to make their widths and heights are equal; step 64, registering the low-resolution reconstructed images with same image sizes with HE staining WSIs corresponding thereto; step 65, cropping the registered low-resolution reconstructed images and the registered HE-staining WSIs according to a size of 512*512 pixels to obtain low-resolution reconstructed image patches and HE staining WSI patches required for network training; step 66, training a mapping relationship between the low-resolution reconstructed image patches and the HE staining WSI patches in step 65 by using a convolutional neural network to obtain the pre-trained super-resolution model trained by the training strategy based on registration mapping; wherein the training strategy of modeling mapping is obtained by the following steps: step 71, cropping any HE staining WSI into HE staining WSI patches with a same size as a computing microscopic image captured by the CMOS image sensor, and using the HE staining WSI patches a

Assignees

• Univ Xidian

Inventors

• Chen Xueli
• Kang huan
• Xie Hui
• Chen Duofang
• Ren Shenghan
• Zhou Wangting

Classifications

• G06T2207/30024

  Cell structures in vitro; Tissue sections in vitro · CPC title

• G06T2207/20192

  Edge enhancement; Edge preservation · CPC title

• G06T2207/20132

  Image cropping · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title