CN113658199B - Regression correction-based chromosome instance segmentation network - Google Patents
Regression correction-based chromosome instance segmentation network Download PDFInfo
- Publication number
- CN113658199B CN113658199B CN202111029543.2A CN202111029543A CN113658199B CN 113658199 B CN113658199 B CN 113658199B CN 202111029543 A CN202111029543 A CN 202111029543A CN 113658199 B CN113658199 B CN 113658199B
- Authority
- CN
- China
- Prior art keywords
- mask
- iou
- segmentation
- chromosome
- regression
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 210000000349 chromosome Anatomy 0.000 title claims abstract description 76
- 230000011218 segmentation Effects 0.000 title claims abstract description 67
- 238000012937 correction Methods 0.000 title claims abstract description 13
- 238000000034 method Methods 0.000 claims abstract description 27
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 8
- 230000001629 suppression Effects 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 4
- 238000013461 design Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000005484 gravity Effects 0.000 claims 1
- 238000001514 detection method Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 7
- 238000003709 image segmentation Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000002679 ablation Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 208000026350 Inborn Genetic disease Diseases 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 208000016361 genetic disease Diseases 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 210000003765 sex chromosome Anatomy 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 206010008805 Chromosomal abnormalities Diseases 0.000 description 1
- 208000031404 Chromosome Aberrations Diseases 0.000 description 1
- 206010010356 Congenital anomaly Diseases 0.000 description 1
- 210000001766 X chromosome Anatomy 0.000 description 1
- 210000002593 Y chromosome Anatomy 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000005452 bending Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000004113 cell culture Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 210000005260 human cell Anatomy 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000016507 interphase Effects 0.000 description 1
- 210000004940 nucleus Anatomy 0.000 description 1
- 238000003793 prenatal diagnosis Methods 0.000 description 1
- 238000001303 quality assessment method Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
Abstract
The invention provides a chromosome instance segmentation network based on regression correction aiming at the problems of low chromosome instance segmentation precision and the like in a real chromosome data set, and can realize high-precision chromosome segmentation. The network corrects the classification confidence through the outputs of the regression branch and the mask branch to obtain higher association with positioning accuracy and segmentation accuracy. A mask-based non-maximum suppression algorithm is provided, and missing segmentation and erroneous segmentation of chromosomes are effectively prevented. A K-IoU loss function is provided that provides more efficient and reasonable loss for the wrong segmentation region. Experimental results show that the method can effectively improve the accuracy of chromosome instance segmentation.
Description
Technical Field
The invention belongs to the field of medical image segmentation, and particularly relates to chromosome image segmentation. Therefore, techniques such as chromosome instance segmentation network based on regression correction are proposed to solve the problem of difficulty in chromosome image segmentation.
Background
Chromosomes are important vectors of genetic information, and normal human cells contain 46 chromosomes, including 22 pairs of autosomes and 1 pair of sex chromosomes (two X sex chromosomes in females and one X chromosome and one Y chromosome in males). Chromosomal abnormalities are an important cause of many congenital genetic diseases, and can occur on each chromosome, including quantitative and morphological variations of the chromosome. Thus, karyotyping of chromosomes is a common and extremely important way of prenatal diagnosis, genetic disease diagnosis and screening. The process of chromosome karyotyping mainly comprises cell culture, shooting imaging, image segmentation and then analyzing, comparing, sorting and classifying chromosomes one by one. The accuracy of chromosome image segmentation directly determines the accuracy and reliability of subsequent chromosome classification and anomaly detection. However, even if the chromosomes are the same number, the chromosomes are different in bending form in the nuclei at different times, and aggregation occurs due to contact and overlapping of the chromosomes as a flexible substance. Currently, the segmentation of overlapping chromosomes is mostly done manually by cytologists, and is highly dependent on the experience of operators, time-consuming, labor-consuming and prone to error.
Chromosome segmentation is used as a primary task of karyotyping, and the accuracy and reliability of subsequent chromosome classification and anomaly detection are directly determined. Traditional manual chromosome segmentation methods rely heavily on manual work, requiring operators to be professionally trained and personnel with a fundamental theory of chromosome processing. In this way, sharma et al distribute the data set to various mass-wrapping platforms by means of mass-wrapping, and enable non-professional personnel to complete manual chromosome segmentation through quality assessment and control methods, and then summarize to complete subsequent classification and anomaly identification. However, this method solves the problem of the number of manpower, but it is difficult to improve the accuracy of segmentation.
The traditional chromosome automatic segmentation method is mostly realized based on geometric morphology, and segmentation of the overlapped chromosomes is realized by extracting the concave points, the tangential points, the thinned skeleton and other characteristics of the overlapped chromosomes. Somasu ndaram et al developed an automated geometric separation method called Multi-object GeodesicActiveContour, MOGAC, which first separated out individual chromosomes. And for the overlapped chromosome, identifying a cutting point on the image through a curvature function, drawing a hypothetical line on the overlapped region by using the obtained cutting point, and finally dividing the overlapped chromosome. YIlmaz et al first performs binarization processing on an input picture, performs morphological sealing processing for a plurality of times to ensure that inter-phase cells keep smooth and round shapes, eliminates noise and unwanted objects, derives distance transformation in a binary image, processes negative values in the distance transformation by using a gray level reconstruction method, adopts a watershed segmentation method to separate single chromosomes and chromosome clusters, and finally finds a geodesic path between the chromosomes in the remaining clusters, and segments overlapping chromosomes by combining contour tangent points calculated by a curvature function. Minaee et al first extract the contours of the overlapping chromosomes, then apply both VAMD (Variations intheAngle ofMotionDirection) and SDTP (SumofDistances amongTotalPoints) criteria to extract the intersection points for dividing the segmentation area, and finally complete the segmentation. This approach allows for simple chromosome cluster segmentation, but has poor segmentation effect for fully overlapping chromosome clusters. The method for dividing the intersection point of the heavy part of the chromosome based on curvature determination has the problems of misjudgment and missed judgment of effective pits, and generally has weaker generalization capability and high accuracy.
With the development of deep learning in recent years, more and more work has been applied to tasks of medical image processing. The chromosome segmentation method based on deep learning can be mainly divided into semantic segmentation and instance segmentation. In the semantic segmentation task, hu et al construct a UNet network containing only two layers of pooling to achieve segmentation of overlapping chromosomes, considering the smaller size of the input image. The segmentation Accuracy (Accuracy) of the overlapping region is high, but the IoU score still needs to be improved; hariyanti et al believe that the addition of pooling and convolution operations in the network facilitates the extraction of more input feature information, and therefore constructing a three-layer UNet network achieves segmentation of overlapping chromosomes with improved segmentation accuracy compared to the result of Hu with IoU scores. However, the potential assumption for the two approaches described above is that there is only a pairwise overlap of chromosomes, and under this assumption the dataset is artificially constructed, but the case of overlapping of chromosomes is much more complex than pairwise overlapping. Therefore, the two methods described above are less viable to apply to real chromosome datasets. In the example segmentation task, bai et al first segments the foreground and background in the real chromosome dataset using U-Net, then obtains the prediction frame of the chromosome from the foreground using YOLOv3, and finally segments the individual chromosome from the prediction frame using UNet. The YOLOv3 adopted in the method has weak detection capability on small targets and overlapped targets, is not suitable for chromosome data sets with serious overlapping conditions, and disassembles an instance segmentation task into three networks for implementation, so that the process is complex. Finally, the accuracy of detection is compared with other data sets, and the reliability is low. Therefore, it is necessary to study an end-to-end instance segmentation method for real chromosome datasets.
Analysis of the dataset revealed that the chromosome was present as a flexible substance in a variety of shapes in the microscopic image. In addition, a large amount of aggregation phenomenon exists in the data set, so that the detection and segmentation difficulty of the chromosome is greatly increased. In general, when detecting the aggregated targets, the classification confidence of the target frame is high, but the actual detection result is poor, so that the AP score at the high IoU threshold is reduced. For this problem, jiang et al predicts IoU the regression and real boxes for replacing the classification confidence to construct IoUNet. Screening errors caused by misleading classification confidence can be eliminated, and the AP score is improved. Wu et al similarly predicts the IoU score of the regression and real boxes and multiplies them with the classification confidence for correcting the classification confidence to construct IoU-aware single-stage object detector. The confidence coefficient has stronger correlation with the positioning precision, and the positioning precision is improved. Chen et al also predicts IoU scores of regression and real boxes to correct classification confidence to improve accuracy of detection of clustered objects, and adds a head network while predicting edges of masks to construct an edge attention monitor network (Supervised EdgeAttentionNetwork, SEANet), improving MaskAP scores at high IoU thresholds. For the case that the classification confidence is high but the actual segmentation result is poor in the example segmentation task, huang et al multiplies IoU of the prediction Mask and the true Mask by the classification confidence to construct Mask Scoring R-CNN (MS RCNN). The method gives consideration to the classification score and the quality score of the prediction mask, and the segmentation result is further improved compared with the mask Rcnn. The above approach allows for correcting classification confidence with IoU of the predicted and real frames or IoU of the predicted and real instances, however, does not consider whether the prediction process is interpretable.
In summary, the existing chromosome segmentation method and confidence correction concept have certain limitations in application scenes, segmentation precision, rationality of training process and the like.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention fully considers the rationality and the relativity of the network prediction process, predicts two kinds of confidence degrees which are more relevant to the positioning precision and the segmentation precision by utilizing the Head branch structure which is more relevant to correct the classification confidence degree, designs a chromosome instance segmentation network based on regression correction, and improves the instance segmentation precision in a real chromosome dataset.
The technical idea of the invention is as follows: the invention uses mask RCNN as a baseline model and Resnet101 as a backbone network, predicts two types of correlation with positioning precision and segmentation precision in a regression branch and a mask branch respectivelyConfidence P of (1) Box And IoU Mask Better correction of classification confidence is achieved. The non-maximum suppression algorithm for screening based on the instance overlapping degree is designed and is more suitable for instance segmentation tasks. A K-IoU loss function is designed to provide more efficient and reasonable loss for the wrong segmentation region. The method effectively solves the missing segmentation and the wrong segmentation of the chromosome instance, and improves the segmentation accuracy.
The implementation scheme comprises the following steps:
(1) Preprocessing the chromosome image;
(1a) Scaling the chromosome image to 512 x 512;
(1b) And generating a corresponding category label.
(2) Predicting regression confidence;
(2a) Processing the loss of prediction box offset in regression branches as regression confidence P Box Is a true value of (2);
(2b) The output of the regression branch is used as the input of the prediction regression confidence coefficient, and the regression confidence coefficient P is obtained through the full connection layer Box 。
(3) Predicting the confidence of the mask;
(3a) IoU using prediction mask and true mask as mask confidence IoU Mask Is a true value of (2);
(3b) The masked confidence IoU is obtained through the full connection layer using the output of the masked branch as an input to the prediction mask confidence Mask 。
(4) Designing a mask-based non-maximum suppression algorithm;
(4a) Reserving the prediction frame as much as possible, and calculating IoU scores of each mask and other masks;
(4b) Other prediction masks with IoU scores greater than the threshold with the current mask are removed by traversing from high to low classification confidence.
(5) Design of a K-IoU loss function;
(5a) Dividing the segmentation result into K parts, and respectively calculating IoU loss in each part;
(5b) The IoU loss weighted sum of each part is calculated using the proportion of the divided area to the total area in each part as a weight.
(6) Training a chromosome instance segmentation network based on regression correction;
(6a) 1/5 of the data set is divided into a test set, 1/5 is divided into a verification set, and the rest is taken as a training set. In view of the small number of real datasets, to prevent overfitting, datasets are first amplified by flipping, rotating. Considering that the image size is large, the training batch is set to 1, and training of 100 epochs is performed using a random gradient descent (SGD), the initial learning rate is 1e-5, the learning momentum is 0.9, and the weight decay is 0.0001.
(6b) Ablation experiment comparison table, as shown in table 2;
(6c) P pair P Box And IoU Mask Exponentiation, comparing the two APs under different exponentiations M Scoring, obtained AP M The scores are shown in table 3.
Compared with the prior art, the invention has the following advantages:
1. the invention realizes higher chromosome segmentation accuracy.
2. The regression confidence coefficient provided by the invention can effectively correct the classification confidence coefficient, thereby reducing missing detection and false detection of the prediction frame.
3. The non-maximum suppression algorithm based on the mask provided by the invention can reduce missing segmentation and error segmentation of the segmentation example.
4. The K-IoU loss function provided by the invention can provide more effective and reasonable loss for error segmentation.
Drawings
Fig. 1 is a model structure of a regression correction network.
FIG. 2 is IoU Box And IoU Mask Comparison graph.
FIG. 3 is a graph comparing IoU loss function to K-IoU loss function.
Detailed description of the preferred embodiments
The invention will be described in further detail with reference to the accompanying drawings and the specific embodiments
Step 1, preprocessing a chromosome image;
(1a) Scaling the chromosome image to 512 x 512, the shorter sides filled with zeros;
(1b) And generating a corresponding category label.
Step 2, predicting regression confidence;
(2a) Calculating the loss of the predicted frame offset in the regression branch, and performing 1-T (L) on the calculated loss by adopting a Smooth L1 loss function reg ) Treatment as regression confidence P Box Of (2), wherein L reg Is the loss of predicted box offset in the regression branch, T (·) is the tanh function;
(2b) Using the offset of the prediction frame in the regression branch as the input of the prediction confidence coefficient, and passing through the full connection layer with 1024 output node numbers and the regression confidence coefficient P in the step (2 a) Box The true value calculation loss of (1) is adopted to obtain regression confidence coefficient P by adopting a cross entropy loss function Box As shown in fig. 1 (a).
Step 3, predicting the confidence of the mask;
(3a) Calculating a loss of the prediction mask in the mask branch, employing a binarized cross entropy loss function, and calculating IoU of the prediction mask and the true mask as mask confidence IoU Mask Is a true value of (2);
(3b) Using the prediction mask outputted by the mask branch as the input of the confidence coefficient of the prediction mask, calculating the loss through the full connection layer with the number of output nodes being 1024 and IoU of the prediction mask and the true mask in the step (3 a), and obtaining the confidence coefficient IoU of the mask by adopting a cross entropy loss function Mask As shown in fig. 1 (b).
Step 4, designing a mask-based non-maximum suppression algorithm;
(4a) Comparing the overlapping condition of the prediction frame and the segmentation result in the chromosome example, as shown in fig. 2;
(4b) Setting the threshold of the original non-maximum suppression algorithm to 0.9, reserving prediction frames as much as possible, and calculating IoU scores of each prediction mask and other prediction masks;
(4c) Other prediction masks with IoU scores greater than the threshold with the current mask are removed by traversing from high to low classification confidence.
Designing a loss function in the steps 5,K-IoU;
(5a) The K-IoU loss function can be expressed asAs shown in fig. 3, the segmentation result is divided into K parts, the invention is designed to be 4, and four parts are obtained by equally dividing two perpendicular central lines;
(5b)δ i represents the proportion of the real mask area in each portion to the entire real mask area IoU i A IoU score representing the prediction mask in each portion and the corresponding true mask is calculated to calculate a IoU penalty weighted sum for each portion.
Step 6, training a chromosome instance segmentation network based on regression correction;
(6a) 1/5 of the data set is divided into a test set, 1/5 is divided into a verification set, and the rest is taken as a training set. In view of the small number of real datasets, to prevent overfitting, datasets are first amplified by flipping, rotating. Considering that the image size is large, only one image is set in the training batch (i.e. one GPU processes one image), and 100epoch training is performed using random gradient descent (SGD), the initial learning rate is 1e-5, the learning momentum is 0.9, and the weight decay is 0.0001.
(6b) Ablation experiment comparison table, as shown in table 2;
(6c) P pair P Box And IoU Mask Performing power operation and multiplying, and multiplying with classification confidence coefficient, namelyComparing the two APs at different indices M Scoring, obtained AP M The scores are shown in table 3.
The experimental environment of the invention is configured as follows: the computer processor is Intel (R) Xeon (R) W-2175CPU@2.50GHz,64GB running memory, NVIDIAGeForce RTX 2080Ti GPU, keras framework. The technical effects of the invention are further described in conjunction with simulation tests as follows:
table 1 comparison of different network model performances
Table 2 ablation experimental design
TABLE 3 comparison of different confidence weight performance
In summary, the invention provides a chromosome instance segmentation network model based on regression correction, which realizes high-precision chromosome segmentation. Experimental results show that compared with the RCNN of the base line network Mask, the method can effectively improve the chromosome instance segmentation accuracy.
Claims (3)
1. A chromosome instance segmentation method based on regression correction is characterized in that a mask RCNN is taken as a baseline model, resnet101 is taken as a backbone network to extract characteristics, and two kinds of confidence coefficient P which are more relevant to positioning precision and segmentation precision are respectively predicted in a regression branch and a mask branch Box And IoU Mask Comprising:
(1) Preprocessing the chromosome image;
(1a) Scaling the chromosome image to 512 x 512;
(1b) Generating a corresponding category label;
(2) Predicting regression confidence;
(2a) Calculating the loss of the predicted frame offset in the regression branch, and performing 1-T (L) on the calculated loss by adopting a Smooth L1 loss function reg ) Treatment as regression confidence P Box True value of (1), whichMiddle L reg Is the loss of predicted box offset in the regression branch, T (·) is the tanh function;
(2b) Using the offset of the prediction frame in the regression branch as the input of the prediction confidence coefficient, and passing through the full connection layer with 1024 output node numbers and the regression confidence coefficient P in the step (2 a) Box The true value calculation loss of (1) is adopted to obtain regression confidence coefficient P by adopting a cross entropy loss function Box ；
(3) Predicting the confidence of the mask;
(3a) Calculating a loss of the prediction mask in the mask branch, employing a binarized cross entropy loss function, and calculating IoU of the prediction mask and the true mask as mask confidence IoU Mask Is a true value of (2);
(3b) The prediction mask outputted by the mask branch is used as the input of the confidence coefficient of the prediction mask, the loss is calculated by the full connection layer with 1024 output node numbers and IoU of the prediction mask and the true mask in (3 a), and the cross entropy loss function is adopted to obtain the confidence coefficient IoU of the mask Mask ；
(4) Designing a mask-based non-maximum suppression algorithm;
(4a) Setting the threshold of the original non-maximum suppression algorithm to 0.9, and calculating IoU scores of each mask and other masks;
(4b) Traversing from high to low according to the classification confidence, and removing other prediction masks with IoU scores greater than a threshold value from the current mask;
(5) K-IoU loss function design;
(5a) Dividing the segmentation result into K parts, and respectively calculating IoU loss of each part;
(5b) The proportion delta of the total area of each part of the divided area i Calculating a IoU loss weighted sum of each part as a weight, and losing as K-IoU;
(6) Training a chromosome instance segmentation method based on regression correction;
(6a) 1/5 of the data set is divided into a test set, 1/5 is divided into a verification set, the rest is taken as a training set, the data set is amplified by means of overturning and rotating in order to prevent overfitting, the training batch is set to be 1 in consideration of large image size, 100epoch training is performed by using random gradient descent, the initial learning rate is 1e-5, the learning momentum is 0.9, and the weight attenuation is 0.0001.
2. The method of claim 1, wherein the chromosome image in step (1 a) is scaled to 512 x 512, the shorter sides being filled with zero values.
3. The method of claim 1, wherein the segmentation result in step (5 a) is divided into K parts, k=4, and four parts are equally divided by two perpendicular centerlines, δ in step (5 b) i Representing the specific gravity of the real mask area in each part to the whole real mask area, the K-IoU loss function is expressed asIoU i IoU score representing the prediction mask in each section and the corresponding real mask. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111029543.2A CN113658199B (en) | 2021-09-02 | 2021-09-02 | Regression correction-based chromosome instance segmentation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111029543.2A CN113658199B (en) | 2021-09-02 | 2021-09-02 | Regression correction-based chromosome instance segmentation network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113658199A CN113658199A (en) | 2021-11-16 |
| CN113658199B true CN113658199B (en) | 2023-11-03 |
Family
ID=78482748
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111029543.2A Active CN113658199B (en) | 2021-09-02 | 2021-09-02 | Regression correction-based chromosome instance segmentation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113658199B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116129123B (en) * | 2023-02-27 | 2024-01-05 | 中国矿业大学 | End-to-end chromosome segmentation method based on uncertainty calibration and region decomposition |
| CN117274294A (en) * | 2023-09-18 | 2023-12-22 | 笑纳科技(苏州)有限公司 | Homologous chromosome segmentation method |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712118A (en) * | 2018-12-11 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of substation isolating-switch detection recognition method based on Mask RCNN |
| CN110533684A (en) * | 2019-08-22 | 2019-12-03 | 杭州德适生物科技有限公司 | A kind of karyotype image cutting method |
| CN110599448A (en) * | 2019-07-31 | 2019-12-20 | 浙江工业大学 | Migratory learning lung lesion tissue detection system based on MaskScoring R-CNN network |
| CN110895810A (en) * | 2019-10-24 | 2020-03-20 | 中科院广州电子技术有限公司 | Chromosome image example segmentation method and device based on improved Mask RCNN |
| CN111062964A (en) * | 2019-11-28 | 2020-04-24 | 深圳市华尊科技股份有限公司 | Image segmentation method and related device |
| CN111160527A (en) * | 2019-12-27 | 2020-05-15 | 歌尔股份有限公司 | Target identification method and device based on MASK RCNN network model |
| CN112036447A (en) * | 2020-08-11 | 2020-12-04 | 复旦大学 | Zero-sample target detection system and learnable semantic and fixed semantic fusion method |
| CN112101083A (en) * | 2019-06-17 | 2020-12-18 | 辉达公司 | Object detection with weak supervision using one or more neural networks |
| WO2020253121A1 (en) * | 2019-06-17 | 2020-12-24 | 商汤集团有限公司 | Target detection method and apparatus, intelligent driving method and device, and storage medium |
| CN112288706A (en) * | 2020-10-27 | 2021-01-29 | 武汉大学 | Automatic chromosome karyotype analysis and abnormality detection method |
| CN112365482A (en) * | 2020-11-16 | 2021-02-12 | 上海北昂医药科技股份有限公司 | Crossed chromosome image example segmentation method based on chromosome trisection feature point positioning |
| CN112949635A (en) * | 2021-03-12 | 2021-06-11 | 北京理工大学 | Target detection method based on feature enhancement and IoU perception |
| CN113033303A (en) * | 2021-02-09 | 2021-06-25 | 北京工业大学 | Method for realizing SAR image rotating ship detection based on RCIoU loss |
| CN113223614A (en) * | 2021-05-31 | 2021-08-06 | 上海澜澈生物科技有限公司 | Chromosome karyotype analysis method, system, terminal device and storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110298298B (en) * | 2019-06-26 | 2022-03-08 | 北京市商汤科技开发有限公司 | Target detection and target detection network training method, device and equipment |
| TWI765262B (en) * | 2019-06-26 | 2022-05-21 | 長佳智能股份有限公司 | Method for training a separation model for separation of overlapping chromosome based on simulation, and method and system for implementing separation of overlapping chromosome using the separation model |
-
2021
- 2021-09-02 CN CN202111029543.2A patent/CN113658199B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712118A (en) * | 2018-12-11 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of substation isolating-switch detection recognition method based on Mask RCNN |
| WO2020253121A1 (en) * | 2019-06-17 | 2020-12-24 | 商汤集团有限公司 | Target detection method and apparatus, intelligent driving method and device, and storage medium |
| CN112101083A (en) * | 2019-06-17 | 2020-12-18 | 辉达公司 | Object detection with weak supervision using one or more neural networks |
| CN110599448A (en) * | 2019-07-31 | 2019-12-20 | 浙江工业大学 | Migratory learning lung lesion tissue detection system based on MaskScoring R-CNN network |
| CN110533684A (en) * | 2019-08-22 | 2019-12-03 | 杭州德适生物科技有限公司 | A kind of karyotype image cutting method |
| CN110895810A (en) * | 2019-10-24 | 2020-03-20 | 中科院广州电子技术有限公司 | Chromosome image example segmentation method and device based on improved Mask RCNN |
| CN111062964A (en) * | 2019-11-28 | 2020-04-24 | 深圳市华尊科技股份有限公司 | Image segmentation method and related device |
| CN111160527A (en) * | 2019-12-27 | 2020-05-15 | 歌尔股份有限公司 | Target identification method and device based on MASK RCNN network model |
| CN112036447A (en) * | 2020-08-11 | 2020-12-04 | 复旦大学 | Zero-sample target detection system and learnable semantic and fixed semantic fusion method |
| CN112288706A (en) * | 2020-10-27 | 2021-01-29 | 武汉大学 | Automatic chromosome karyotype analysis and abnormality detection method |
| CN112365482A (en) * | 2020-11-16 | 2021-02-12 | 上海北昂医药科技股份有限公司 | Crossed chromosome image example segmentation method based on chromosome trisection feature point positioning |
| CN113033303A (en) * | 2021-02-09 | 2021-06-25 | 北京工业大学 | Method for realizing SAR image rotating ship detection based on RCIoU loss |
| CN112949635A (en) * | 2021-03-12 | 2021-06-11 | 北京理工大学 | Target detection method based on feature enhancement and IoU perception |
| CN113223614A (en) * | 2021-05-31 | 2021-08-06 | 上海澜澈生物科技有限公司 | Chromosome karyotype analysis method, system, terminal device and storage medium |
Non-Patent Citations (10)
| Title |
|---|
| A Clinical Dataset and Various Baselines for Chromosome Instance Segmentation;Runhua Huang 等;IEEE/ACM Transactions on Computational Biology and Bioinformatics;1-9 * |
| ARMS Net: Overlapping chromosome segmentation based on Adaptive Receptive field Multi-Scale network;Guangjie Wang等;Biomedical Signal Processing and Control;第68卷;102811 * |
| Chromosome Extraction Based on U-Net and YOLOv3;Bai, H 等;IEEE Access;第8卷;178563-178569 * |
| IoU-aware single-stage object detector for accurate localization;Shengkai Wu等;Image and Vision Computing;第97卷;103911 * |
| Joint Object Contour Points and Semantics for Instance Segmentation;Wenchao Zhang等;arXiv:2008.00460v3 [cs.CV];1-11 * |
| Mask Scoring R-CNN;Huang, Z 等;32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition. CVPR;6402–6411 * |
| Supervised Edge Attention Network for Accurate Image Instance Segmentation;Chen, X等;Proceeding of 20th European Conference on Computer Vision;617–631 * |
| 基于RNA-Seq的长非编码RNA预测;孙磊 等;生物化学与生物物理进展;第39卷(第12期);1156-1166 * |
| 基于改进的Mask R-CNN的染色体图像分割框架;冯涛 等;计算机应用;第40卷(第11期);3332-3339 * |
| 基于深度学习的数字病理图像识别分析与应用研究;孙沐毅;中国博士学位论文全文数据库 (医药卫生科技辑)(第（2021）01期);E059-43 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113658199A (en) | 2021-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109117876B (en) | Dense small target detection model construction method, dense small target detection model and dense small target detection method | |
| JP6710135B2 (en) | Cell image automatic analysis method and system | |
| CN109584209B (en) | Vascular wall plaque recognition apparatus, system, method, and storage medium | |
| CN109492706B (en) | Chromosome classification prediction device based on recurrent neural network | |
| CN113658199B (en) | Regression correction-based chromosome instance segmentation network | |
| CN111612008A (en) | Image segmentation method based on convolution network | |
| US20060204953A1 (en) | Method and apparatus for automated analysis of biological specimen | |
| Fujita et al. | Cell detection and segmentation in microscopy images with improved mask R-CNN | |
| CN110705403A (en) | Cell sorting method, cell sorting device, cell sorting medium, and electronic apparatus | |
| CN112819821B (en) | Cell nucleus image detection method | |
| CN112132827A (en) | Pathological image processing method and device, electronic equipment and readable storage medium | |
| Liu et al. | SRAS‐net: Low‐resolution chromosome image classification based on deep learning | |
| Chen et al. | Segmentation of overlapping cervical cells with mask region convolutional neural network | |
| CN113343755A (en) | System and method for classifying red blood cells in red blood cell image | |
| CN113177927A (en) | Bone marrow cell classification and identification method and system based on multiple features and multiple classifiers | |
| CN111598854A (en) | Complex texture small defect segmentation method based on rich robust convolution characteristic model | |
| CN115359264A (en) | Intensive distribution adhesion cell deep learning identification method | |
| CN111931763A (en) | Depth scene text detection method based on random shape edge geometric modeling | |
| CN114972202A (en) | Ki67 pathological cell rapid detection and counting method based on lightweight neural network | |
| CN113221731B (en) | Multi-scale remote sensing image target detection method and system | |
| Gajjar et al. | Intersection over Union based analysis of Image detection/segmentation using CNN model | |
| CN112580624B (en) | Multi-directional text region detection method and device based on boundary prediction | |
| Nan et al. | Automatic fine-grained glomerular lesion recognition in kidney pathology | |
| Pijackova et al. | Deep learning pipeline for chromosome segmentation | |
| Peng et al. | Hers superpixels: Deep affinity learning for hierarchical entropy rate segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |