CN110782451A - Suspected microcalcification area automatic positioning method based on discriminant depth confidence network - Google Patents
Suspected microcalcification area automatic positioning method based on discriminant depth confidence network Download PDFInfo
- Publication number
- CN110782451A CN110782451A CN201911065975.1A CN201911065975A CN110782451A CN 110782451 A CN110782451 A CN 110782451A CN 201911065975 A CN201911065975 A CN 201911065975A CN 110782451 A CN110782451 A CN 110782451A
- Authority
- CN
- China
- Prior art keywords
- image
- mammary gland
- ddbns
- microcalcification
- area
- 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.)
- Granted
Links
- 208000004434 Calcinosis Diseases 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 34
- 210000005075 mammary gland Anatomy 0.000 claims abstract description 42
- 210000000481 breast Anatomy 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000012545 processing Methods 0.000 claims abstract description 21
- 238000001514 detection method Methods 0.000 claims abstract description 19
- 230000002308 calcification Effects 0.000 claims abstract description 15
- 238000007781 pre-processing Methods 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 11
- 230000002708 enhancing effect Effects 0.000 claims abstract description 3
- 230000009467 reduction Effects 0.000 claims abstract description 3
- 230000000877 morphologic effect Effects 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 8
- 238000000354 decomposition reaction Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000012937 correction Methods 0.000 claims description 5
- 238000012216 screening Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000011049 filling Methods 0.000 claims description 4
- 210000002976 pectoralis muscle Anatomy 0.000 claims description 4
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000008030 elimination Effects 0.000 claims description 2
- 238000003379 elimination reaction Methods 0.000 claims description 2
- 230000003902 lesion Effects 0.000 abstract description 3
- 238000003745 diagnosis Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 9
- 206010006187 Breast cancer Diseases 0.000 description 7
- 208000026310 Breast neoplasm Diseases 0.000 description 7
- 238000009499 grossing Methods 0.000 description 4
- 238000009607 mammography Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 201000011510 cancer Diseases 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 210000004907 gland Anatomy 0.000 description 1
- 210000002364 input neuron Anatomy 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000004630 mental health Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10116—X-ray image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30068—Mammography; Breast
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Informatics (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a suspected microcalcification area automatic positioning method based on a discriminant depth confidence network, which comprises the following steps: 1) preprocessing a mammary gland X-ray image: segmenting and enhancing the mammary gland region; 2) sample acquisition and pretreatment: segmenting the enhanced mammary gland image to obtain a subblock image set for model training, and performing noise reduction and background removal processing on the subblocks; 3) sub-block feature extraction and classification: constructing a Discriminant Deep Belief Network (DDBNs), training and fine-tuning a DDBNs model, and completing feature extraction and automatic classification of the breast sub-blocks; 4) detection of microcalcification area: inputting a mammary gland X-ray image to be detected, applying a trained optimal model after a series of preprocessing on the image, classifying and distinguishing the subblocks, and marking a suspicious microcalcification region according to a distinguishing result. The invention can complete the automatic detection and positioning of suspicious lesion areas, effectively reduces the false positive rate while obtaining higher detection rate, and the detected calcification area has high consistency with the expert marking area.
Description
Technical Field
The invention belongs to the technical field of medical image processing, and particularly relates to a suspected microcalcification area automatic positioning method based on a discriminant depth confidence network.
Background
Breast cancer is one of the common malignant tumors threatening the physical and mental health of women. Microcalcification clusters in mammographic images are an important sign of early breast cancer, and research data suggest that: between 30% and 50% of breast cancer patients diagnosed with microcalcification cluster in early mammography screening. The calcification diagnosis based on the mammography examination is the first choice for breast cancer diagnosis and screening nowadays, but because the contrast between micro-calcifications and surrounding tissues of some early-stage breast cancers is very low, human eyes are difficult to identify, missed diagnosis is often caused, and the early-stage diagnosis largely depends on radiologists, so the influence of subjective factors is large. Therefore, in the general breast examination, an auxiliary method is urgently needed to help a doctor to make a quick and accurate diagnosis in a short time, improve the reading efficiency and the diagnosis accuracy of a radiologist, and reduce missed diagnosis and misdiagnosis. The method of processing images by using a computer assists doctors in early breast cancer diagnosis, and becomes one of research hotspots and difficult problems of breast image processing. In the implementation of the computer aided detection and diagnosis of breast cancer, the first important link is how to find and automatically locate the suspicious lesion region of the breast image.
The detection algorithm for the calcified breast areas mainly comprises methods based on morphology, wavelet analysis, mixed features, artificial neural networks and the like, a large number of features need to be designed and extracted manually, and the feature extraction process is large in calculation amount and time-consuming. At present, the deep learning method shows a good application prospect in the aspect of lesion detection of medical images, overcomes the limitation of shallow learning compared with the traditional manual design of feature description, does not need manual design to extract features, can extract more expressive and richer essential features from a data set, and improves the accuracy of classification.
Because the microcalcifications in the mammographic image are very small, different in size, different in shape, and variable in distribution, the detection of the microcalcifications in the mammographic image still does not achieve a satisfactory effect. Therefore, further discussion and research is needed to help clinicians locate microcalcification regions in mammographic images more quickly and accurately, improve microcalcification detection rate, and reduce false detection rate.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a method for automatically positioning a suspected micro-calcification area based on a discriminant deep belief network, which effectively reduces the false positive rate while ensuring higher detection rate, and the detected calcification area has high consistency with an expert marking area.
In order to achieve the above purpose, the method for automatically positioning the suspected microcalcification region based on the discriminant deep belief network of the present invention comprises the following steps:
1): preprocessing a mammary gland X-ray image;
(1) breast region segmentation
The images used in the present invention are all from the mammography screening database DDSM. Firstly, smoothing an image by adopting a 9X9 mean filter, carrying out binarization processing on the image by using a global threshold technology (threshold TH1 is 0.1) to obtain a mask, and removing label interference by adopting a maximum connected region algorithm; then projecting the obtained mask on an original image, smoothing the image again by using a 39x39 mean filter, and performing global threshold processing on the smoothed image by using a threshold (the threshold TH2 is 0.5) to remove pectoral muscle interference; and finally, calculating the minimum occupying area of the mammary gland, and realizing the extraction of the mammary gland area.
(2) Mammary region enhancement
Firstly, morphological filtering is carried out on the extracted mammary gland region image, a flat disc structural element with the radius of 5 is used, and the result of morphological contrast enhancement is that the result of top-hat transformation of the image is added to the original mammary gland region image, and then the result of bottom-hat transformation of the image is subtracted. And then, gamma conversion is adopted to reduce the image contrast of the low gray value area and improve the image contrast of the high gray value area.
2): obtaining and preprocessing a sample;
(1) sample acquisition
And obtaining sample data for DDBNs training and testing. The enhanced breast region image was segmented, overlapping into sub-blocks of size 96x96 (the small blocks overlap 75%, primarily to increase the number of samples containing microcalcification cluster small blocks). The sub-blocks are divided into three categories: one is a negative sample, containing no normal small pieces of calcification; one is a positive sample, a small piece containing microcalcifications; one type is the mammary boundary patch.
(2) Pretreatment of subblocks
And decomposing and reconstructing the breast subblocks by adopting a 'sym 4' wavelet basis function, setting the number of decomposition layers to be 5, setting approximate coefficients obtained after wavelet decomposition to be zero, and carrying out hard threshold processing on detail coefficients of 1-5 layers. And taking the value obtained by adding the standard deviation to the mean value of the horizontal high-frequency coefficient of each layer as the threshold of the layer, carrying out hard threshold processing on the horizontal, vertical and diagonal high-frequency coefficients of the layer, and then carrying out wavelet reconstruction. And finally, further removing the background and the small non-calcified independent points by adopting a threshold value and area removing method, and then performing gray value linear stretching.
3): extracting and classifying the sub-block characteristics;
(1) DDBNs model construction and training
The DDBNs network is formed by stacking three layers of RBMs, wherein the first two layers of the model adopt GRBMs to extract the morphological characteristics of breast subblocks, the third layer adopts DRBMs with characteristic learning and classification capabilities, pre-training is carried out according to the extracted morphological characteristics, and DDBNs initialization is completed by adopting weight parameters obtained by the pre-training.
(2) DDBNs model fine tuning
After the layer-by-layer pre-training of the DDBNs is finished, the DDBNs are converted into a deep neural network which is supervised and classified by using a Softmax regression layer, network parameters are adjusted by a network through a minimized loss function and back propagation, and the automatic classification of the breast subblocks is finished.
4): detecting microcalcification areas;
firstly, preprocessing an original mammary gland X-ray image in the step 1), and then dividing an enhanced mammary gland region image into small blocks with the same size of 96X96 (the small blocks are divided according to the overlapping condition of 50 percent, and if the mammary gland region image is not divided completely, boundary zero filling processing is required to be firstly carried out); then, all the small blocks obtained after the division are subjected to small block pretreatment in the step 2); and finally, traversing and classifying and distinguishing the small blocks by applying the trained optimal model in the step 3), and if the small blocks are judged to be small blocks containing calcifications and contain at least two calcifications, marking the positions of the small blocks on the mammary gland image to finally finish the marking of all suspicious microcalcification area.
The invention has the following beneficial effects:
the automatic suspected microcalcification region positioning method based on the discriminant depth confidence network adopts an image enhancement algorithm combining top cap, low cap transformation and Gamma correction, and the constructed DDBNs network has better deep feature mining and classifying capability. Through experimental verification of 105 mammary X-ray images containing microcalcifications in a mammography screening database (DDSM) of south Florida State university, the method effectively reduces the false positive rate (2.17%) while ensuring a high detection rate (99.45%), the detected calcifications area is highly consistent with the expert marking area, the running time for detecting a 1831X 4021 image is only 15s probably, and the requirement of rapidity is met. The method can better help doctors to make quick and accurate diagnosis in a short time, improves the reading efficiency and the diagnosis accuracy of radiologists, and provides a new research idea for detecting the microcalcification clusters in the mammary gland X-ray images.
Drawings
The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, together with the embodiments of the invention. In the drawings:
fig. 1 is a general block diagram of the present invention.
Fig. 2 is a diagram of a breast region extraction process in an embodiment of the present invention.
Fig. 3 is a diagram of a breast area enhancement process in an embodiment of the present invention.
Fig. 4 is a diagram of a breast sub-block pretreatment process in an embodiment of the present invention.
FIG. 5 shows the structure of the DDBNs model constructed in the invention.
Fig. 6 is a diagram of case detection results in an embodiment of the invention.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.
As shown in fig. 1, the method for automatically locating a suspected microcalcification region based on a discriminant deep belief network according to the present invention includes the following steps: 1) preprocessing a mammary gland X-ray image: segmenting and enhancing the mammary gland region; 2) sample acquisition and pretreatment: segmenting the enhanced mammary gland image to obtain a subblock image set for model training, and performing noise reduction and background removal processing on the subblocks; 3) sub-block feature extraction and classification: constructing a Discriminant Deep Belief Network (DDBNs), and performing feature extraction and automatic classification on the breast subblocks; 4) detection of microcalcification area: inputting a mammary gland X-ray image to be detected, applying a trained optimal model after a series of preprocessing on the image, classifying and distinguishing the subblocks, and marking a suspicious microcalcification region according to a distinguishing result.
1): preprocessing a mammary gland X-ray image;
(1) breast region segmentation
Firstly, smoothing an image by adopting a 9X9 mean filter, carrying out binarization processing on the image by using a global threshold technology (threshold TH1 is 0.1) to obtain a mask, and removing label interference by adopting a maximum connected region algorithm; then projecting the obtained mask on an original image, smoothing the image again by using a 39x39 mean filter, and performing global threshold processing on the smoothed image by using a threshold (the threshold TH2 is 0.5) to remove pectoral muscle interference; and finally, calculating the minimum occupying area of the mammary gland, and realizing the extraction of the mammary gland area.
The process of extracting the mammary gland region is shown in fig. 2: where a is an original image, b is a binary image, c is a binary mask image with labels removed, d is an original image with labels removed, e is a image with pectoral muscles removed, and f is a breast region image.
(2) Mammary region enhancement
Firstly, morphological filtering is carried out on the extracted mammary gland region image, a flat disc structural element with the radius of 5 is used, and the result of morphological contrast enhancement is shown as a formula (1):
I＝(I
P+I
τ)-I
B(1)
i in the above formula
pAs an original image, I
TFor top-hat transformation results, I
BIs the bottom-cap transformation result.
Then, the morphologically enhanced image is subjected to Gamma conversion, and the formula of Gamma correction is shown in (2):
f＝cr
γ(2)
in the formula (2), r is a gray scale value of the original image, c and γ are constants (c is 1 and γ is 2), and f is a gray scale value of the image after gamma conversion.
The mammary gland region enhancement process diagram is shown in fig. 3: a is a diagram of a protomammary gland region, b is a diagram of morphological contrast enhancement, and c is a diagram after Gamma correction.
2): obtaining and preprocessing a sample;
(1) sample acquisition
And obtaining sample data for DDBNs training and testing. The enhanced breast region image was segmented, overlapping into sub-blocks of size 96x96 (the small blocks overlap 75%, primarily to increase the number of samples containing microcalcification cluster small blocks). The sub-blocks are divided into three categories: one is a negative sample, containing no normal small pieces of calcification; one is a positive sample, a small piece containing microcalcifications; one type is the mammary boundary patch. The data set was screened for a total of 6000 patches per type.
(2) Pretreatment of subblocks
Wavelet transform has the capability of multi-resolution analysis. And decomposing and reconstructing the breast subblocks by adopting a 'sym 4' wavelet basis function, setting the number of decomposition layers to be 5, setting approximate coefficients obtained after wavelet decomposition to be zero, and carrying out hard threshold processing on detail coefficients of 1-5 layers. And taking the value obtained by adding the standard deviation to the mean value of the horizontal high-frequency coefficient of each layer as the threshold of the layer, carrying out hard threshold processing on the horizontal, vertical and diagonal high-frequency coefficients of the layer, and then carrying out wavelet reconstruction.
In order to further remove the background, a threshold method is adopted for processing, and pixels with the gray value less than 15 are set to be zero; in order to remove the small non-calcified independent points, an area removal method is adopted to remove objects with the area smaller than 16 in the small images; and finally, performing gray value linear stretching.
The breast sub-block pretreatment process is illustrated in fig. 4: a is a patch containing microcalcifications, b is a graph after wavelet decomposition and reconstruction, c is a graph after threshold and area elimination, and d is a graph after linear stretching.
3): extracting and classifying the sub-block characteristics;
(1) DDBNs model construction and training
The DDBNs model constructed by the invention is shown in fig. 5, a DDBNs network is formed by stacking three layers of RBMs, the first two layers of the model adopt GRBMs to extract the morphological characteristics of breast subblocks, the third layer adopts DRBMs with characteristic learning and classification capabilities, pre-training is carried out according to the extracted morphological characteristics, and DDBNs initialization is completed by adopting weight parameters obtained by the pre-training. Adopting a DDBNs-3 structure: 9216-1000-1000-1000-3.
The training of the RBM for constructing DDBNs is carried out by maximizing the likelihood of training samples, as shown in formula (3):
to theta
1,2Logarithmic probability derivation
，Demand for
Conditional probability distribution of
In which the joint probability distribution of
But instead of the other end of the tube
If the calculation is not good, a comprehensive algorithm of continuous free energy contrast divergence (FEPCD) and a fast learning algorithm (CD) needs to be adopted, namely the GRBM of the first layer is trained by adopting the FEPCD algorithm, and the GRBM of the second layer and the DRBM of the top layer are trained by adopting the CD algorithm.
Setting parameters: batch training size: 100, respectively; learning rate: 0.1; total training times: 50; momentum: [0.5,0.4,0.3,0.2,0.1,0].
The data set is from the small blocks after the pretreatment of the sub-blocks, the number of three types of samples in the training set is 5000 respectively, and the number of three types of samples in the testing set is 1000 respectively.
Firstly, the unlabelled sample data normalized according to the minimum maximum value is sent into the GRBM of the first two layers of DDBNs, the network parameters are obtained through unsupervised training, and the morphological characteristics of the breast sample are obtained. Then, the network parameters are sent into a top-level DRBM of DDBNs, and are subjected to supervised learning to obtain posterior probability of the breast sample under each category, so that classification of the breast sample is completed.
(2) DDBNs model fine tuning
After the layer-by-layer pre-training of DDBNs is completed, it is converted into a deep neural network that uses a Softmax regression layer for supervised classification. In order to solve parameter redundancy, a loss function is adopted to add attenuation penalty items to realize a parameter penalty mode. And (3) by using Softmax, adjusting network parameters by reversely propagating the network through a minimum formula (4) loss function, and finishing automatic classification of the breast subblocks.
Wherein θ ∈ R
n+1Is the model parameter of Softmax, m is the trainingNumber of samples, k number of sample class, 1 (-) as an indicator function, representing if the term is true, 1, else 0, y sample class label, x
iIs the input sample characteristic of the top layer, and n is the number of input neurons of the top layer.
Setting parameters: batch training size: 100, respectively; total training times: 300, respectively; learning rate: 1; momentum: 0.5.
(3) DDBNs Performance analysis
The following 3 criteria were used: true Positive Rate (TPR), False Positive Rate (FPR), and Overall Accuracy (OA). Initialization, training and the like of the deep learning network have certain randomness, so that the classification results of different times have deviation. The accuracy and stability of the classification of the DDBNs based breast samples was therefore assessed by 5 experiments on the data set, calculating OA using statistical methods of mean and standard deviation. The experimental results are shown in table 1 below.
TABLE 1 results of the experiment
As can be seen from Table 1, the DDBNs model constructed by the invention achieves about 99.2% of detection rate and about 2.7% of false positive rate, the overall accuracy is about 98.3%, and the false positive rate is effectively reduced while the high detection rate is ensured.
4): detecting microcalcification areas;
firstly, preprocessing an original mammary gland X-ray image in the step 1), and then dividing an enhanced mammary gland region image into small blocks with the same size of 96X96 (the small blocks are divided according to the overlapping condition of 50 percent, and if the mammary gland region image is not divided completely, boundary zero filling processing is required to be firstly carried out); then, all the small blocks obtained after the division are subjected to small block pretreatment in the step 2); and finally, traversing and classifying and distinguishing the small blocks by applying the trained optimal model in the step 3), and if the small blocks are judged to be small blocks containing calcifications and contain at least two calcifications, marking the positions of the small blocks on the mammary gland image to finally finish the marking of all suspicious microcalcification area.
There is a restriction on the traversal of the small blocks: in order to reduce the operand and the program running time, only all the non-zero small blocks after preprocessing are traversed.
A Case of cancer _04Case A _1096_1.LEFT _ CC in the DDSM database was analyzed and compared with the expert labeling results, and the final results are shown in Table 2 and FIG. 6. In fig. 6: a is an expert marked area map and b is a detection map of the method of the invention.
Table 2 case test results
The method detects 105 images in the DDSM database, and contains 662 calcification point clusters, so that the method obtains 99.45% of detection rate and 2.17% of false detection rate; the detected calcification area has high consistency with the expert marking area; and the running time for detecting an image with the size of 1831X 4021 is only 15s approximately, so that the requirement on rapidity is met.
Finally, it should be noted that: while embodiments of the invention have been disclosed above, it is not limited to the applications listed in the description and the embodiments, which are fully applicable in all kinds of fields of application of the invention, and further modifications may readily be effected by those skilled in the art, so that the invention is not limited to the specific details without departing from the general concept defined by the claims and the scope of equivalents.
Claims (5)
1. The method for automatically positioning the suspected microcalcification area based on the discriminant deep belief network is characterized by comprising the following steps of:
1) preprocessing a mammary gland X-ray image: inputting an original mammary gland X-ray image, and segmenting and enhancing a mammary gland region;
2) sample acquisition and pretreatment: segmenting the enhanced mammary gland image obtained in the step 1) to obtain a subblock image set, dividing subblocks into three types, and performing noise reduction, background removal and other processing on the subblocks;
3) sub-block feature extraction and classification: constructing a Discriminant Deep Belief Network (DDBNs), normalizing the sample data set preprocessed in the step 2), and then sending the normalized sample data set into the DDBNs network for training and fine tuning to complete the feature extraction and automatic classification of subblocks;
4) detection of microcalcification area: inputting a mammary gland X-ray image to be detected, applying the trained optimal model in the step 3) after a series of preprocessing is carried out on the original image, carrying out classification and judgment on the subblocks, and marking the suspicious microcalcification region according to a judgment result.
2. The method for automatically locating the suspected microcalcification area based on the discriminant deep belief network as claimed in claim 1, wherein: the step 1) comprises the following processes:
firstly, removing interference items such as pectoralis muscles and labels in a mammary gland image by using algorithms such as an average filter, a global threshold, a maximum connected region, mask operation, a minimum circumscribed rectangle and the like, and realizing extraction of a mammary gland region;
then, the mammary gland area is enhanced by combining morphological top hat, low hat transformation and Gamma correction, and a flat disc structural element with the radius of 5 is used, and the result of morphological contrast enhancement is shown as the formula (1):
I＝(I
P+I
T)-I
B(1)
wherein I
pAs an original image, I
TFor top-hat transformation results, I
BIs the bottom-cap transformation result;
the formula of Gamma correction is shown in (2):
f＝cr
γ(2)
where r is the gray scale value of the original image, c and γ are constants (c is 1 and γ is 2), and f is the gray scale value of the image after gamma conversion.
3. The method for automatically locating the suspected microcalcification area based on the discriminant deep belief network as claimed in claim 1, wherein: the step 2) comprises the following processes:
sample acquisition: dividing the enhanced mammary gland region image into small blocks with the same size of 96x96, and if the mammary gland region image is not divided completely, performing boundary zero filling treatment; dividing sub-blocks, screening a data set, and dividing the small blocks into three types, namely small blocks containing micro-calcifications, normal small blocks without calcifications and small breast boundary blocks;
pretreatment of small blocks: carrying out denoising processing on the small blocks by using wavelet analysis, decomposing and reconstructing breast sub-blocks by using 'sym 4' wavelet basis functions, setting the number of decomposition layers to be 5, setting approximate coefficients obtained after wavelet decomposition to be zero, and carrying out hard threshold processing on detail coefficients of 1-5 layers: and taking the mean value plus the standard deviation value of each layer of horizontal high-frequency coefficients as the threshold value of the layer, carrying out hard threshold processing on the horizontal, vertical and diagonal high-frequency coefficients of the layer, then carrying out wavelet reconstruction, and further removing the background and the small non-calcified independent points by adopting a threshold value method and an area elimination method.
4. The method for automatically locating the suspected microcalcification area based on the discriminant deep belief network as claimed in claim 1, wherein: the step 3) comprises the following steps:
constructing and training a DDBNs model: the DDBNs are composed of two layers of generation limited Boltzmann machines (GRBM) and one layer of discriminant limited Boltzmann machine (DRBM); adopting a DDBNs-3 structure: 9216 1000-3; the RBM training for constructing DDBNs is carried out by maximizing the likelihood of training samples, the GRBM of the first layer is trained by adopting an FEPCD algorithm, and the GRBM of the second layer and the DRBM of the top layer are trained by adopting a CD algorithm;
fine tuning of a DDBNs model: after the layer-by-layer pre-training of the DDBNs is finished, the DDBNs are converted into a deep neural network which is supervised and classified by using a Softmax regression layer, network parameters are adjusted by the network through a minimum loss function and back propagation, and the automatic classification of the breast subblocks is finished.
5. The method for automatically locating the suspected microcalcification area based on the discriminant deep belief network as claimed in claim 1, wherein: in step 4), firstly, the original mammary gland X-ray image is preprocessed in the step 1), and then the enhanced mammary gland area image is divided into small blocks with the same size of 96X96 (the small blocks are divided according to the overlapping condition of 50%, and if the mammary gland area image is not divided completely, boundary zero filling processing is needed to be firstly carried out); then, carrying out denoising and background removing treatment in the step 2) on all the small blocks obtained after segmentation; and finally, traversing and classifying and distinguishing the small blocks by applying the trained optimal model in the step 3), and if the small blocks are judged to be small blocks containing calcifications and contain at least two calcifications, marking the positions of the small blocks on the mammary gland image to finally finish the marking of all suspicious microcalcification area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911065975.1A CN110782451B (en) | 2019-11-04 | 2019-11-04 | Suspected microcalcification area automatic positioning method based on discriminant depth confidence network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911065975.1A CN110782451B (en) | 2019-11-04 | 2019-11-04 | Suspected microcalcification area automatic positioning method based on discriminant depth confidence network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110782451A true CN110782451A (en) | 2020-02-11 |
| CN110782451B CN110782451B (en) | 2021-09-07 |
Family
ID=69388830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911065975.1A Expired - Fee Related CN110782451B (en) | 2019-11-04 | 2019-11-04 | Suspected microcalcification area automatic positioning method based on discriminant depth confidence network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110782451B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325266A (en) * | 2020-02-18 | 2020-06-23 | 慧影医疗科技（北京）有限公司 | Method and device for detecting micro-calcified clusters in breast molybdenum target image and electronic equipment |
| US20230047497A1 (en) * | 2019-06-10 | 2023-02-16 | Obshchestvo S Ogranichennoj Otvetstvennost'yu 'medicinskie Skrining Sistemy | System for processing radiographic images and outputting the result to a user |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010046499A1 (en) * | 2008-10-24 | 2010-04-29 | Fenics | Image processing method comprising the display of regions of interest in an x-ray image according to scores associated with said regions |
| US20100104155A1 (en) * | 2007-08-06 | 2010-04-29 | Shoupu Chen | Method for detection of linear structures and microcalcifications in mammographic images |
| CN101853376A (en) * | 2010-02-10 | 2010-10-06 | 西安理工大学 | Computer aided detection method for microcalcification in mammograms |
| CN107392204A (en) * | 2017-07-20 | 2017-11-24 | 东北大学 | A kind of galactophore image microcalcifications automatic checkout system and method |
| CN109829896A (en) * | 2019-01-14 | 2019-05-31 | 中国科学院苏州生物医学工程技术研究所 | The micro-calcification clusters automatic testing method of digital galactophore tomography X image based on multi-domain characteristics |
-
2019
- 2019-11-04 CN CN201911065975.1A patent/CN110782451B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100104155A1 (en) * | 2007-08-06 | 2010-04-29 | Shoupu Chen | Method for detection of linear structures and microcalcifications in mammographic images |
| WO2010046499A1 (en) * | 2008-10-24 | 2010-04-29 | Fenics | Image processing method comprising the display of regions of interest in an x-ray image according to scores associated with said regions |
| CN101853376A (en) * | 2010-02-10 | 2010-10-06 | 西安理工大学 | Computer aided detection method for microcalcification in mammograms |
| CN107392204A (en) * | 2017-07-20 | 2017-11-24 | 东北大学 | A kind of galactophore image microcalcifications automatic checkout system and method |
| CN109829896A (en) * | 2019-01-14 | 2019-05-31 | 中国科学院苏州生物医学工程技术研究所 | The micro-calcification clusters automatic testing method of digital galactophore tomography X image based on multi-domain characteristics |
Non-Patent Citations (1)
| Title |
|---|
| 高洁 等: "基于微钙化点检测的乳腺X光片计算机辅助诊断系统", 《哈尔滨理工大学学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230047497A1 (en) * | 2019-06-10 | 2023-02-16 | Obshchestvo S Ogranichennoj Otvetstvennost'yu 'medicinskie Skrining Sistemy | System for processing radiographic images and outputting the result to a user |
| CN111325266A (en) * | 2020-02-18 | 2020-06-23 | 慧影医疗科技（北京）有限公司 | Method and device for detecting micro-calcified clusters in breast molybdenum target image and electronic equipment |
| CN111325266B (en) * | 2020-02-18 | 2023-07-21 | 慧影医疗科技(北京)股份有限公司 | Detection method and device for microcalcification clusters in breast molybdenum target image and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110782451B (en) | 2021-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mahmood et al. | A brief survey on breast cancer diagnostic with deep learning schemes using multi-image modalities | |
| CN107067402B (en) | Medical image processing apparatus and breast image processing method thereof | |
| Pharwaha et al. | Shannon and non-shannon measures of entropy for statistical texture feature extraction in digitized mammograms | |
| Liu et al. | Microcalcification detection in full-field digital mammograms with PFCM clustering and weighted SVM-based method | |
| Phadke et al. | Fusion of local and global features for classification of abnormality in mammograms | |
| Nusantara et al. | Classification of digital mammogram based on nearest-neighbor method for breast cancer detection | |
| Fan et al. | Mass detection and segmentation in digital breast tomosynthesis using 3D-mask region-based convolutional neural network: a comparative analysis | |
| CN110782451B (en) | Suspected microcalcification area automatic positioning method based on discriminant depth confidence network | |
| Beheshti et al. | Classification of abnormalities in mammograms by new asymmetric fractal features | |
| Nedra et al. | Detection and classification of the breast abnormalities in Digital Mammograms via Linear Support Vector Machine | |
| Ganvir et al. | Filtering method for pre-processing mammogram images for breast cancer detection | |
| Hikmah et al. | An image processing framework for breast cancer detection using multi-view mammographic images | |
| Sharma et al. | Roi segmentation using local binary image | |
| Krishnaveni et al. | Study of Mammogram Microcalcification to aid tumour detection using Naive Bayes Classifier | |
| CN110136112B (en) | Computer-aided detection algorithm based on mammary gland X-ray photography calcification | |
| Dong et al. | Feature selection and analysis on mammogram classification | |
| Zyout et al. | Empirical mode decomposition of digital mammograms for the statistical based characterization of architectural distortion | |
| Marrocco et al. | Detection of cluster of microcalcifications based on watershed segmentation algorithm | |
| ABUBACKER et al. | AN IMPROVED PERIPHERAL ENHANCEMENT OF MAMMOGRAM IMAGES BY USING FILTERED REGION GROWING SEGMENTATION. | |
| Singh et al. | Segmentation and characterization of breast tumor in mammograms | |
| Zyout et al. | A new approach for the detection of architectural distortions using textural analysis of surrounding tissue | |
| Domingues | An automatic mammogram system: from screening to diagnosis | |
| Boujelben et al. | Automatic application level set approach in detection calcifications in mammographic image | |
| Sharma et al. | Dual Thresholding based Breast cancer detection in Mammograms | |
| Dabass et al. | Segmentation of Noisy Mammograms Using Hybrid Techniques |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210907 |