EP-4170583-B1 - METHOD AND DEVICE FOR DETECTING CIRCULATING ABNORMAL CELLS

EP4170583B1EP 4170583 B1EP4170583 B1EP 4170583B1EP-4170583-B1

Inventors

FAN, Xianjun
LAN, Xingjie
YE, XIN
ZHANG, YI
LI, Congsheng

Dates

Publication Date: 20260506
Application Date: 20201204

Claims (9)

A method for detecting circulating abnormal cells, comprising: segmenting and labelling cell nuclei included in dark field microscope images of a plurality of probe channels respectively, by using an image processing algorithm and a morphological algorithm; inputting the dark field microscope images, in which the cell nuclei are labelled, of the plurality of probe channels into a pre-built circulating abnormal cell detection model to acquire a number of staining signals included in each of the labelled cell nuclei in the dark field microscope image of each of the probe channels; and for each of the labelled cell nuclei, determining whether the labelled cell nucleus belongs to a circulating abnormal cell, based on the acquired number of the staining signals included in the labelled cell nucleus in the dark field microscope image of each of the probe channels.
The method according to claim 1, wherein building the circulating abnormal cell detection model comprises: segmenting and labelling the cell nuclei included in the dark field microscope sample images of the plurality of probe channels respectively by using the image processing algorithm and the morphological algorithm, and segmenting the dark field microscope sample images, in which the cell nuclei are labelled, of the probe channels to acquire a plurality of cell nucleus sample images; for each of the cell nucleus sample images, performing multiple convolution processes on the cell nucleus sample image to respectively acquire a first feature image, a second feature image, a third feature image, and a fourth feature image; performing a convolution process on the fourth feature image to acquire a fifth feature image, and performing upsampling on the fifth feature image to acquire a sixth feature image; performing a convolution process on the third feature image and the sixth feature image to acquire a seventh feature image, and performing upsampling on the seventh feature image to acquire an eighth feature image; performing a convolution process on the second feature image and the eighth feature image to acquire a ninth feature image, and performing upsampling on the ninth feature image to acquire a tenth feature image; performing a convolution process on the first feature image and the tenth feature image to acquire an eleventh feature image; and training and testing a deep learning network by taking each of the cell nucleus sample images as an input of the deep learning network, fusing the seventh feature image, the ninth feature image, and the eleventh feature image as output prediction results in three scales of the deep learning network, and finally taking the number of the staining signals included in the labelled cell nuclei in the cell nucleus sample images as the output of the deep learning network, to acquire a circulating abnormal cell detection model.
The method according to claim 2, wherein performing the multiple convolution processes on the cell nucleus sample image to respectively acquire the first feature image, the second feature image, the third feature image, and the fourth feature image comprises: sequentially performing convolution processes of a first convolution layer, a second convolution layer, a third convolution layer, and a fourth convolution layer on the cell nucleus sample image to acquire the first feature image; performing convolution processes of a fifth convolution layer and a sixth convolution layer on the first feature image to acquire the second feature image; performing convolution processes of a seventh convolution layer and an eighth convolution layer on the second feature image to acquire the third feature image; and performing convolution processes of a ninth convolution layer and a tenth convolution layer on the third feature image to acquire the fourth feature image.
The method according to claim 3, wherein in sequentially performing the convolution processes of the first convolution layer, the second convolution layer, the third convolution layer, and the fourth convolution layer on the cell nucleus sample image to acquire the first feature image: a size of the cell nucleus sample image is 320*320, a number of convolution kernels of the first convolution layer is 32, a size of the convolution kernels is 3*3 and a step size is 1, and a size of the output feature image is 320*320; the second convolution layer comprises a first convolution sublayer and a second convolution sublayer, wherein a number of convolution kernels of the first convolution sublayer is 32, a size of the convolution kernels is 3*3 and a step size is 2, and a size of the output feature image is 160* 160; and a number of convolution kernels of the second convolution sublayer is 64, a size of the convolution kernels is 1*1 and a step size is 1, and a size of the output feature image is 160* 160; the third convolution layer comprises a third convolution sublayer and a fourth convolution sublayer, wherein a number of convolution kernels of the third convolution sublayer is 64, a size of the convolution kernels is 3*3 and a step size is 1, and a size of the output feature image is 160* 160; and a number of convolution kernels of the fourth convolution sublayer is 128, a size of the convolution kernels is 1*1 and a step size is 2, and a size of the output feature image is 80*80; and the fourth convolution layer comprises a fifth convolution sublayer and a sixth convolution sublayer, wherein a number of convolution kernels of the fifth convolution sublayer is 128, a size of the convolution kernel is 3*3 and a step size is 1, and a size of the output feature image is 80*80; and a number of convolution kernels of the sixth convolution sublayer is 128, a size of the convolution kernels is 1*1 and a step size is 1, and a size of the output feature image is 80*80, the feature image output by the sixth convolution sublayer being the first feature image.
The method according to any one of claims 1 to 4, wherein segmenting and labelling the cell nuclei included in the dark field microscope images of the plurality of probe channels respectively by using the image processing algorithm and the morphological algorithm comprises: for the dark field microscope image of each of the probe channels, performing Gaussian kernel filtering on the dark field microscope image to acquire a denoised image; labelling connected domains of the denoised image to acquire labelled connected domains; and segmenting the acquired connected domains by using the morphological algorithm, and labelling the segmented domains to acquire labelled cell nuclei.
The method according to any one of claims 1 to 4, wherein inputting the dark field microscope images, in which the cell nuclei are labelled, of the plurality of probe channels into the pre-built circulating abnormal cell detection model to acquire the number of staining signals included in each of the labelled cell nuclei in the dark field microscope image of each of the probe channels comprises: inputting a dark field microscope image in which the cell nuclei are labelled of a first probe channel into the circulating abnormal cell detection model to acquire a first count of staining signals included in each of the labelled cell nuclei in the dark field microscope image, in which the cell nuclei are labelled, of the first probe channel; inputting a dark field microscope image in which the cell nuclei are labelled of a second probe channel into the circulating abnormal cell detection model to acquire a second count of staining signals included in each of the labelled cell nuclei in the dark field microscope image, in which the cell nuclei are labelled, of the second probe channel; inputting a dark field microscope image in which the cell nuclei are labelled of a third probe channel into the circulating abnormal cell detection model to acquire a third count of staining signals included in each of the labelled cell nuclei in the dark field microscope image, in which the cell nuclei are labelled, of the third probe channel; and inputting a dark field microscope image in which the cell nuclei are labelled of a fourth probe channel into the circulating abnormal cell detection model to acquire a fourth count of staining signals included in each of the labelled cell nuclei in the dark field microscope image, in which the cell nuclei are labelled, of the fourth probe channel.
The method according to claim 6, wherein for each of the labelled cell nuclei, determining whether the labelled cell nucleus belongs to a circulating abnormal cell based on the acquired number of the staining signals included in the labelled cell nucleus in the dark field microscope image of each of the probe channels comprises: acquiring a first count of staining signals included in a first labelled cell nucleus in the dark field microscope image, in which the cell nuclei are labelled, of the first probe channel; acquiring a second count of staining signals included in the first labelled cell nucleus in a dark field microscope image, in which the cell nuclei are labelled, of a second probe channel; acquiring a third count of staining signals included in the first labelled cell nucleus in a dark field microscope image, in which the cell nuclei are labelled, of a third probe channel; acquiring a fourth count of staining signals included in the first labelled cell nucleus in a dark field microscope image, in which the cell nuclei are labelled, of a fourth probe channel; and determining whether the first labelled cell nucleus belongs to a circulating abnormal cell based on the first count, the second count, the third count, and the fourth count of the staining signals.
An electronic device, comprising: a memory, a processor, and a bus, wherein the memory stores machine readable instructions executable by the processor; when the electronic device is operating, the processor communicates with the memory through the bus; and when the machine readable instructions are executed by the processor, the steps of the method for detecting circulating abnormal cells according to any one of claims 1 to 7 are performed.
A computer readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs steps of the method for detecting circulating abnormal cells according to any one of claims 1 to 7.

Description

CROSS-REFERENCE TO RELATED APPLICATION The present disclosure claims the benefit of a priority of Chinese Patent Application No. 202010585145.8 and titled "METHOD AND DEVICE FOR DETECTING CIRCULATING ABNORMAL CELLS", filed with the China National Intellectual Property Administration on June 23, 2020. TECHNICAL FIELD The present disclosure relates to the technical field of cell detection, in particular to a method and a device for detecting circulating abnormal cells (CACs). BACKGROUND Circulating tumor cells (CTCs) are cells that are shed from primary tumors and form secondary tumors in distant organ sites during tumor metastasis. A large number of studies have shown that the number of CTCs in blood can predict the disease progression and indicate the response of the tumor to chemotherapy drugs. Therefore, by collecting a certain amount of peripheral blood, detecting CTCs in the peripheral blood and monitoring changes of CTCs' content in the blood, it is possible to analyze the pathogenesis of tumors and evaluate the prognosis of patients, and thus understand the improvement of patients' clinical status and drug resistance after receiving treatment. In recent years, circulating genetically abnormal cells (CACs) have been reported to be found in the peripheral blood of patients with non-small cell lung cancer (NSCLC). These cells are of a type considered to be involved in the occurrence, progression, and metastasis of lung cancer, including CTCs that are shed from the tumor and enter the peripheral blood circulation system. The detection of CACs in peripheral blood makes it possible to predict the existence of tumor earlier. This method undoubtedly has a broad prospect for application. At present, the method of detecting CACs is to collect the dark field (DF) microscope images of peripheral blood, label CACs in the DF microscope images based on the morphological information of CACs such as shape and/or size and based on manual intervention, count the labelled CACs, and determine the content of CACs in blood based on the counting results. However, this method for detecting CACs is based on the morphological information of CACs in combination with manual intervention to detect, determine, and count CACs, so the detection is rather subjective and not reliable. Also, the detection efficiency is low and the detection cost is high due to the need for manual participation. R. L. KATZ ET AL: "Genetically Abnormal Circulating Cells in Lung Cancer Patients: An Antigen-Independent Fluorescence In situ Hybridization-Based Case-Control Study",CLINICAL CANCER RESEARCH, vol. 16, no. 15, 1 August 2010 (2010-08-01), pages 3976-3987, discloses detecting CACs using image processing and manual review by an experienced observer. SUMMARY In view of the foregoing, an object of the present disclosure is to provide a method and a device for detecting circulating abnormal cells to improve the reliability of detection of circulating abnormal cells. In a first aspect, an embodiment of the present disclosure provides a method for detecting circulating abnormal cells, comprising: segmenting and labelling cell nuclei included in dark field microscope images of a plurality of probe channels respectively, by using an image processing algorithm and a morphological algorithm;inputting the dark field microscope images, in which the cell nuclei are labelled, of the plurality of probe channels into a pre-built circulating abnormal cell detection model to acquire the number of staining signals included in each of the labelled cell nuclei in the dark field microscope image of each of the probe channels; andfor each of the labelled cell nuclei, determining whether the labelled cell nucleus belongs to a circulating abnormal cell, based on the acquired number of the staining signals included in the labelled cell nucleus in the dark field microscope image of each of the probe channels. With reference to the first aspect, an embodiment of the present disclosure provides a first possible implementation of the first aspect, wherein building the circulating abnormal cell detection model comprises: segmenting and labelling cell nuclei included in dark field microscope sample images of a plurality of probe channels respectively by using the image processing algorithm and the morphological algorithm, and segmenting the dark field microscope sample images, in which the cell nuclei are labelled, of the probe channels to acquire a plurality of cell nucleus sample images;for each of the cell nucleus sample images, performing a convolution process several times on the cell nucleus sample image to acquire a first feature image, a second feature image, a third feature image and a fourth feature image, respectively;performing a convolution process on the fourth feature image to acquire a fifth feature image, and performing upsampling on the fifth feature image to acquire a sixth feature image;performing a convolution process on the third feature image and the sixth