CN113299369B - Medical image window adjusting optimization method - Google Patents
Medical image window adjusting optimization method Download PDFInfo
- Publication number
- CN113299369B CN113299369B CN202110526220.8A CN202110526220A CN113299369B CN 113299369 B CN113299369 B CN 113299369B CN 202110526220 A CN202110526220 A CN 202110526220A CN 113299369 B CN113299369 B CN 113299369B
- Authority
- CN
- China
- Prior art keywords
- window
- image
- task
- adjusting
- level
- 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
Images
Classifications
-
- 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 relates to a window adjusting optimization method for a medical image. The method comprises the steps of firstly constructing a window adjusting optimization module, wherein the window adjusting optimization module is constructed by a window setting submodule and a converting submodule. And secondly, constructing a task requirement network, wherein the task requirement network is determined by specific task requirements. Then model training and verification are carried out. And finally, outputting window width and window level parameters according to task requirements, and outputting the window-adjusted and optimized image according to the requirements. According to the invention, the window width and window position parameters can be adaptively optimized by combining task requirements through the window adjusting optimization module, and the window adjusting optimized image can be generated.
Description
Technical Field
The invention relates to the technical field of medical image processing, in particular to a window adjusting optimization method for a medical image.
Background
Dicom (digital Imaging and Communications in medicine), which is an international standard for medical image and related information, is an international standard (ISO 12052) for medical image and related information. DICOM format image pixels often contain rich gray scale information, and the pixel value range is large, generally 4096 gray scales or even higher. The visual display of DICOM medical images is related to the extraction of key and important information by imaging doctors. The medical image window adjusting processing is that on the basis that the gray level of an original image is mapped to a medical high-brightness display, a doctor dynamically adjusts the WL parameter of a window width WW window level, and a symptom point is highlighted in a local range so as to observe and judge symptoms. The window level WL refers to the center of the gray scale value range displayed by the image and can be used for adjusting the brightness of the image; the window width WW is the range of gray scale values displayed by the image, and can be used to adjust the contrast of the image.
In the medical image, each tissue has a relatively stable gray scale range, and a doctor needs to clearly display normal tissues and suspected lesion points in an ROI (region of interest) through a window. When a doctor actually adjusts the window, the doctor generally adjusts the window finely on the basis of the fixed window width and window position according to experience, such as a brain window (WL:50, WW:100) and a bone window (WL:300, WW: 1500).
Different window width window levels are set to highlight image features of DICOM images under different gray levels, the fixed window width window level is not a suitable parameter of an ROI image of a specific tissue, doctors need to slightly adjust the window width window level to improve the significance of symptom points under the specific tissue, however, the window adjusting process is time-consuming and labor-consuming and is limited by the experience of image doctors, and the image display effect is not ideal after manual window adjustment.
With the rapid development and use of deep learning in various fields, the deep learning technology is applied to the window adjusting technology, window width and window level parameters can be automatically obtained around task requirements and used for window adjusting setting of DICOM images, and the significance of disease display is enhanced to the greatest extent.
Disclosure of Invention
The invention aims to provide a DICOM medical image window width and window level optimization method aiming at the problems of time and labor waste and poor window adjusting effect in the window adjusting process of a doctor, generates optimized window width and window level parameters around specific task requirements, and can output window-adjusted images under the task requirements according to the generated optimized window width and window level parameters.
Under different task requirements, the optimized window width and window level parameters are different, such as a brain window and a bone window which respectively highlight two gray level characteristics with larger difference of brain tissues and bones. The method of the invention designs different subsequent task networks aiming at the specific task detection purpose, thereby achieving the purpose of adjusting the window in a targeted optimization way.
The invention provides a window-adjusting optimization method of DICOM medical images based on deep learning, which comprises the following steps:
step 1, constructing a window adjusting optimization module
The window adjusting optimization module is constructed by a window setting submodule and a conversion submodule.
The window setting sub-module consists of 4 layers of ConV and 2 layers of FC, a plurality of convolution layers of the module learn image characteristics, and the last layer of full-connection layer outputs an image window width WW and a window level WL.
The conversion submodule is composed of a window transformation function and is defined as follows:
wherein WW sets the window width of the sub-module output for the window, WL sets the window level of the sub-module output for the window, yminLower limit of gray level, y, mapped for window transform functionmaxAnd (4) the gray level upper limit after the window transformation function mapping.
Step 2, constructing a task demand network
The task requirement network is determined by specific task requirements. If the task target is to classify the input image into a disease state, the task demand network uses a classification network such as Resnet50, Densenet121, etc. And constructing different task demand networks according to different task demands.
Step 3, model training and verification
Firstly, selecting a proper data set, both public and private data sets according to task requirements, randomly dividing the data set into a training set, a verification set and a test set according to a proportion, and scaling the pictures to the same size.
Secondly, model training is carried out on the selected data set by using the window adjusting optimization module and the task requirement network which are well established in the steps 1 and 2. And when the model training is finished, selecting a model evaluation index according to the task requirement and the data set, testing the trained model on the test set, and selecting the model with the highest evaluation index as a final model.
Step 4, outputting window width and window level parameters surrounding the task requirements, and outputting the window-adjusting optimized image according to the requirements
And (4) outputting window width and window level parameters surrounding the task requirement by using the window adjusting optimization module of the final model obtained in the step (3), and outputting the window adjusting optimized image according to the requirement. The DICOM format original image and the detection task are input, a first part of output, namely an optimized window width and window level parameter surrounding the task requirement, is generated through a window sub-setting module, and a second part of output, namely an image after window optimization is adjusted according to the task requirement, is generated through a conversion sub-module according to the generated optimized window width and window level parameter.
The invention has the following beneficial effects: according to the invention, the window width and window position parameters can be adaptively optimized by combining task requirements through the window adjusting optimization module, and the window adjusting optimized image can be generated.
Drawings
FIG. 1 is a block diagram of the window adjustment optimization module of the present invention;
FIG. 2 is a flow chart of an embodiment of the method of the present invention;
fig. 3a to 3f are diagrams illustrating window adjusting effects according to an embodiment of the method of the present invention.
Detailed Description
The following examples are given to further illustrate embodiments of the present invention.
In the embodiment, the generation of the image after the window adjustment for the classification significance of the intracranial bleeding disorder is taken as an example, but the method is not limited to the generation of the image after the window adjustment for the optimization window width of the intracranial bleeding disorder, and can be applied to the optimization window adjustment of various medical images such as X-ray photography, Computed Tomography (CT), Magnetic Resonance Imaging (MRI) and the like.
Referring to fig. 1 and fig. 2, the invention relates to a window adjusting optimization method for medical images, comprising the following steps:
step 1, constructing a window adjusting optimization module
The window adjusting optimization module is constructed by a window setting submodule and a conversion submodule, and is shown in fig. 1.
Wherein, the window setting submodule consists of 4 layers of ConV and 2 layers of FC. And (3) generating parameters through random initialization, learning image characteristics by a plurality of convolution layers of the module, and outputting an image window width WW and a window level WL by a last layer of full connection layer.
The conversion submodule is composed of a window transformation function and is defined as follows:
wherein x is HU value (pixel value) of input DICOM format image, y is gray value (pixel value) converted from DICOM into gray image, WW is window width output by window setting submodule, and WL is window level output by window setting submodule，yminIs the lower limit of the gray level after the window transformation function mapping, y in this embodimentmin＝0，ymaxIs the upper limit of the gray level after the window transformation function mapping, y in this embodimentmax＝255。
Step 2, constructing a task demand network
In the embodiment, the purpose that the output DICOM window-adjusted image can maximally display the significance of intracranial hemorrhage symptoms is taken as a task, and the pretrained IncephetionResNet V2 is taken as a task requirement network and used for extracting the characteristics of the window-adjusted and optimized image generated by the window-adjusting optimization module, so that symptoms are judged, and a classification result is output.
Step 3, model training and verification
First, an appropriate data set is selected according to task requirements, the present embodiment adopts an RSNA intracranial hemorrhage public data set, and randomly divides the data set, wherein 85% of the data is used as a training set, 10% of the data is used as a testing set, and 5% of the data is used as a verification set, and the pictures are scaled to the same size.
Secondly, model training is carried out on the selected data set by using the window adjusting optimization module and the task requirement network which are well established in the steps 1 and 2, and window adjusting optimization module parameters and Inception ResNet V2 network parameters are updated according to loss back propagation. When the model training is finished, because the RSNA data set is used in this embodiment, AUC is selected as a model evaluation index, the model obtained by training is tested on the test set, and the model with the highest average AUC is selected as a final model.
Step 4, taking classification of intracranial bleeding as an example, outputting window width and window level parameters surrounding task requirements, and outputting an image after window adjustment of the optimized window width and window level
And (4) outputting window width and window level parameters surrounding the task requirement by using the window adjusting optimization module of the final model obtained in the step (3), and outputting the window adjusting optimized image according to the requirement. The DICOM format original image and the intracranial hemorrhage classification task are input, a first part of output, namely an optimized window width and window level parameter surrounding a task requirement, is generated through a window sub-setting module, and a second part of output, namely an image after window optimization is adjusted according to the task requirement is generated through a conversion sub-module according to the generated optimized window width and window level parameter.
Fig. 3a to 3f are comparison diagrams of five images of intracranial hemorrhage and a comparison diagram of the window-width-fixed window adjustment effect and the window-width-optimized window adjustment effect of the normal image, specifically, fig. 3a is an image of an intracerebral hemorrhage window, the left diagram in fig. 3a is a fixed window, the right diagram is an optimized window (the lower left diagram is a fixed window, and the right diagram is an optimized window), fig. 3b is an image of a window for regulating intrinsic hemorrhage, fig. 3c is an image of a window for regulating epidural hemorrhage, fig. 3d is an image of a window for regulating subdural hemorrhage, fig. 3e is an image of a window for regulating subarachnoid hemorrhage, and fig. 3f is an image of a window for regulating normal image. Through the legend, the problems of fuzzy bleeding point boundaries, confusion between bleeding areas and normal tissue display and the like of the fixed-window-width window position adjustment window image can be found. After window adjustment and optimization, the outline of the cerebral hemorrhage disease point in the image is clearer, the contrast between the normal tissue and the hemorrhage area is enhanced, so that a doctor can distinguish the normal tissue and the hemorrhage point more easily, and better image support is provided for the diagnosis of the doctor.
Claims (2)
1. A window-adjusting optimization method for medical images is characterized by comprising the following steps:
step 1, constructing a window adjusting optimization module
The window adjusting optimization module consists of a window setting submodule and a conversion submodule;
the window setting submodule is used for learning image characteristics and outputting optimized image window width and window level;
the conversion submodule is used for generating an image after window adjustment and optimization;
step 2, constructing a task demand network
The task requirement network is determined by specific task requirements and is used for extracting the characteristics of the window-adjusted and optimized image generated by the window-adjusting optimization module;
step 3, model training and verification
Firstly, selecting a proper data set according to task requirements, randomly dividing the data set into a training set, a verification set and a test set according to a proportion, and simultaneously scaling pictures to the same size;
secondly, performing model training on the selected data set by using the window adjusting optimization module and the task demand network built in the steps 1 and 2; when the model training is finished, selecting a model evaluation index according to the task requirement and the data set, testing the trained model on a test set, and selecting the model with the highest evaluation index as a final model;
step 4, outputting window width and window level parameters surrounding the task requirements, and outputting the window-adjusting optimized image according to the requirements
Inputting a DICOM format original image and a detection task, generating a first part of output through a window sub-setting module, namely optimizing window width and window level around task requirements, and generating a second part of output according to the generated optimizing window width and window level through a conversion sub-module, namely adjusting a window optimized image under the task requirements;
the conversion submodule is realized by a window transformation function, and the window transformation function is defined as follows:
wherein x is HU value of input DICOM format image, y is gray value converted from DICOM to gray image, WW is window width of window setting submodule output, WL is window level of window setting submodule output, yminLower limit of gray level, y, mapped for window transformation functionmaxAnd (4) the gray level upper limit after the window transformation function mapping.
2. The medical image windowing optimization method according to claim 1, characterized by:
the window setting submodule consists of four convolutional layers and two full-connection layers, wherein the convolutional layers learn the image characteristics, and the last full-connection layer outputs the image window width and the window level.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110526220.8A CN113299369B (en) | 2021-05-14 | 2021-05-14 | Medical image window adjusting optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110526220.8A CN113299369B (en) | 2021-05-14 | 2021-05-14 | Medical image window adjusting optimization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113299369A CN113299369A (en) | 2021-08-24 |
| CN113299369B true CN113299369B (en) | 2022-06-10 |
Family
ID=77321984
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110526220.8A Active CN113299369B (en) | 2021-05-14 | 2021-05-14 | Medical image window adjusting optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113299369B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115439423B (en) * | 2022-08-22 | 2023-09-12 | 北京医准智能科技有限公司 | CT image-based identification method, device, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796036A (en) * | 2019-10-14 | 2020-02-14 | 大连理工大学 | Method for improving identification precision of structural modal parameters |
| CN111833321A (en) * | 2020-07-07 | 2020-10-27 | 杭州电子科技大学 | Window-adjusting optimization-enhanced intracranial hemorrhage detection model and construction method thereof |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7869637B2 (en) * | 2006-07-31 | 2011-01-11 | Siemens Medical Solutions Usa, Inc. | Histogram calculation for auto-windowing of collimated X-ray image |
| KR102037416B1 (en) * | 2012-12-17 | 2019-10-28 | 삼성전자주식회사 | Method for managing of external devices, method for operating of an external device, host device, management server, and the external device |
-
2021
- 2021-05-14 CN CN202110526220.8A patent/CN113299369B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796036A (en) * | 2019-10-14 | 2020-02-14 | 大连理工大学 | Method for improving identification precision of structural modal parameters |
| CN111833321A (en) * | 2020-07-07 | 2020-10-27 | 杭州电子科技大学 | Window-adjusting optimization-enhanced intracranial hemorrhage detection model and construction method thereof |
Non-Patent Citations (1)
| Title |
|---|
| 靳艳敏等.小型通用医学图像处理软件的设计与实现.《医疗卫生装备》.2010,(第01期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113299369A (en) | 2021-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10129553B2 (en) | Dynamic digital image compression based on digital image characteristics | |
| EP3716202A1 (en) | Systems and methods to facilitate review of liver tumor cases | |
| US20150287188A1 (en) | Organ-specific image display | |
| CN103544684A (en) | Image processing method and image processing apparatus | |
| CN111127430A (en) | Method and device for determining medical image display parameters | |
| JP2022517769A (en) | 3D target detection and model training methods, equipment, equipment, storage media and computer programs | |
| CN110288698B (en) | Meniscus three-dimensional reconstruction system based on MRI | |
| CN113299369B (en) | Medical image window adjusting optimization method | |
| CN113689337B (en) | Ultrasonic image super-resolution reconstruction method and system based on generation countermeasure network | |
| CN111918610A (en) | Gradation conversion method for chest X-ray image, gradation conversion program, gradation conversion device, server device, and conversion method | |
| CN112562058A (en) | Rapid establishing method of intracranial vascular simulation three-dimensional model based on transfer learning | |
| US9552663B2 (en) | Method and system for volume rendering of medical images | |
| CN116258933A (en) | Medical image segmentation device based on global information perception | |
| CN113205148B (en) | Medical image frame interpolation method and terminal for iterative interlayer information fusion | |
| CN114519770A (en) | Lumbosacral plexus nerve segmentation and three-dimensional visualization method | |
| CN106683058A (en) | Calibrating method of medical image and device thereof | |
| TW202218627A (en) | Spine measurement and state assessment method including a positioning step, a line drawing step, a measuring step, a selection step and a calculation step | |
| CN112509080A (en) | Method for establishing intracranial vascular simulation three-dimensional model based on transfer learning | |
| CN116912305B (en) | Brain CT image three-dimensional reconstruction method and device based on deep learning | |
| JP7129754B2 (en) | Image processing device | |
| US11354806B2 (en) | Detection of regions with low information content in digital X-ray images | |
| Subramani et al. | Optimal Fuzzy Intensification System for Contrast Distorted Medical Images | |
| Andreini et al. | Multi-stage Synthetic Image Generation for the Semantic Segmentation of Medical Images | |
| Aboobacker et al. | Improving presentation consistency of radiographic images using deep learning | |
| CN114863029A (en) | Bone imaging method based on VR technology |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221214 Address after: Room 806, Building 1, Wuzhou International Business Center, Xihu District, Hangzhou City, Zhejiang Province, 310000 Patentee after: Hangzhou Shuzhilaida Technology Co.,Ltd. Address before: 310018 No.2 street, Baiyang street, Qiantang New District, Hangzhou City, Zhejiang Province Patentee before: HANGZHOU DIANZI University |