CN-121707862-B - Machine vision-assisted diabetes wound surface image noise suppression method and system

Abstract

The application discloses a machine vision-assisted diabetes wound surface image noise suppression method and a machine vision-assisted diabetes wound surface image noise suppression system, which relate to the technical field of medical imaging, and comprise the steps of extracting reflection image features of a reflection area of a diabetes foot wound surface image; analyzing and determining the type of the reflective interference, if the type of the reflective interference is the reflective of the exudates, predicting the calculation error of the wound surface area based on the characteristics of the reflective image, formulating an adaptive image denoising mechanism according to the prediction area error, and performing reflective noise inhibition treatment on the diabetic foot wound surface image. The method solves the technical problem that the existing method for suppressing the image noise of the wound surface of the diabetic foot can not effectively solve the interference caused by the reflection of the area of the wound surface, so that the calculation accuracy of the area of the wound surface is insufficient.

Inventors

• WANG LI
• CUI SHENGNAN
• ZHANG ZHIGANG

Assignees

• 陕西中医药大学

Dates

Publication Date

20260512

Application Date

20260213

Claims (9)

1. 1. The machine vision-assisted diabetes wound surface image noise suppression method is characterized by comprising the following steps of: extracting the reflective image characteristics of the reflective area of the diabetic foot wound surface image through a convolutional neural network; Determining a reflection interference type according to the reflection image feature analysis; If the reflection interference type is exudate reflection, predicting a wound area calculation error based on the reflection image characteristics to obtain a predicted area error; formulating an adaptive image denoising mechanism according to the predicted area error, and performing reflection noise inhibition treatment on the diabetic foot wound surface image; The method for predicting the calculation error of the wound surface area based on the reflection image features, obtaining the prediction area error, comprises the following steps: Based on historical diabetes diagnosis records, taking exudate reflection as constraint, collecting a sample reflection image feature set, counting historical wound surface area calculation errors of different sample reflection image features in a historical detection process as sample area errors, and obtaining a sample area error set; adopting the sample reflection image feature set and the sample area error set as training data, and carrying out K-fold cross division to obtain K sample training sets, wherein K is an integer greater than or equal to 8; respectively training a deep learning model to be converged by using the K sample training sets to generate K area error predictors; And analyzing the area error prediction complexity according to the reflective image features to obtain error prediction complexity, calculating the error prediction of the wound area based on the error prediction complexity, K area error predictors and the reflective image features, and outputting a predicted area error.
2. 2. The machine vision-assisted diabetes wound image noise suppression method according to claim 1, wherein extracting the reflection image features of the reflection region of the diabetes foot wound image through the convolutional neural network comprises: Configuring a reflective image characteristic index, wherein the reflective image characteristic index at least comprises color saturation, hue, texture uniformity, spatial position, brightness gradient characteristics, shape characteristics and size characteristics; based on historical diabetes diagnosis records, collecting a sample reflection area wound surface image set of diabetes, and labeling different sample reflection area wound surface images according to the reflection image characteristic indexes to obtain a sample reflection image characteristic set; the wound surface image set of the sample reflection area and the characteristic set of the sample reflection image are used as training data, a convolutional neural network is trained to converge, and a reflection image characteristic extraction model is constructed; and extracting the reflective image features of the reflective region of the diabetic foot wound surface image by using the reflective image feature extraction model.
3. 3. The machine vision-assisted diabetes wound image noise suppression method according to claim 1, wherein a reflection interference type is determined by performing interference type judgment according to the reflection image characteristics according to a preset characteristic identification rule table, wherein the reflection interference type comprises exudate reflection, granulation tissue reflection and eschar reflection.
4. 4. The machine vision-assisted diabetes wound image noise suppression method according to claim 3, wherein if the reflection interference type is granulation tissue reflection, reflection component separation and recombination are performed on the diabetes foot wound image according to a first image processing strategy, wherein the first image processing strategy is to eliminate highlights by estimating and compressing local illumination components, and simultaneously protect and enhance object inherent reflection components to preserve color and texture characteristics of granulation tissue.
5. 5. The machine vision-assisted diabetes wound image noise suppression method according to claim 3, wherein if the reflective interference type is eschar reflective, edge-aware brightness suppression and boundary preservation processing are performed on the diabetes foot wound image according to a second image processing strategy, wherein the second image processing strategy is to maintain a clear physical structure boundary between eschar and normal tissue while reducing the brightness of a eschar edge reflective area through guided filtering.
6. 6. The machine vision-assisted diabetes wound image noise suppression method according to claim 1, wherein the step of acquiring a sample area error comprises: Collecting an initial wound area of an original diabetic foot wound image which corresponds to the characteristics of the sample reflective image and is not subjected to reflective noise inhibition treatment; Collecting standard wound surface area of a standard diabetic foot wound surface image subjected to reflection noise inhibition treatment, which corresponds to the characteristics of a sample reflection image; taking the ratio of the area difference between the initial wound surface area and the standard wound surface area to the standard wound surface area as a sample area error.
7. 7. The machine vision-assisted diabetes wound image noise suppression method according to claim 6, wherein performing an area error prediction complexity analysis according to the reflex image features to obtain an error prediction complexity, performing a wound area calculation error prediction based on the error prediction complexity, K area error predictors, and reflex image features, and outputting a predicted area error, comprising: Constructing a first feature identification matrix according to the reflective image features; constructing a sample feature identification matrix set according to the sample reflection image feature set; Performing similar traversal comparison on the first feature identification matrix and a plurality of sample feature identification matrices in the sample feature identification matrix set respectively, determining a plurality of sample similarities, and calculating the average value to obtain comprehensive sample similarities; taking the reciprocal of the integrated sample similarity as error prediction complexity; Multiplying the error prediction complexity by P to obtain J, wherein P is the initial selected number of the area error predictors, P is 3, if the calculation result of J is smaller than 1, J is equal to 1, and if the calculation result of J is larger than K, J is equal to K; And randomly selecting J area error predictors from the K area error predictors, respectively carrying out wound area calculation error prediction according to the reflective image characteristics, and carrying out average value calculation on J prediction results to obtain a prediction area error.
8. 8. The machine vision-assisted diabetes wound image noise suppression method according to claim 1, wherein formulating an adaptive image denoising mechanism according to the prediction area error comprises: Pre-constructing an area error-denoising strategy mapping table of exudates reflection, wherein the denoising strategy comprises an image reflection denoising algorithm, and the denoising strength of the image reflection denoising algorithm is positively correlated with the area error; and determining an adaptive image denoising mechanism according to the predicted area error matching by using the area error-denoising strategy mapping table.
9. 9. A machine vision-assisted diabetic wound image noise suppression system for performing the machine vision-assisted diabetic wound image noise suppression method of any one of claims 1-8, comprising: the image feature extraction module is used for extracting the reflective image features of the reflective area of the diabetic foot wound surface image through the convolutional neural network; The interference type judging module is used for determining the reflection interference type according to the reflection image characteristic analysis; The wound surface error prediction module is used for predicting the calculation error of the wound surface area based on the reflection image characteristics to obtain the prediction area error if the reflection interference type is the exudate reflection; the wound surface image processing module is used for formulating an adaptive image denoising mechanism according to the prediction area error and carrying out reflection noise suppression processing on the diabetic foot wound surface image; The method for predicting the calculation error of the wound surface area based on the reflection image features, obtaining the prediction area error, comprises the following steps: Based on historical diabetes diagnosis records, taking exudate reflection as constraint, collecting a sample reflection image feature set, counting historical wound surface area calculation errors of different sample reflection image features in a historical detection process as sample area errors, and obtaining a sample area error set; adopting the sample reflection image feature set and the sample area error set as training data, and carrying out K-fold cross division to obtain K sample training sets, wherein K is an integer greater than or equal to 8; respectively training a deep learning model to be converged by using the K sample training sets to generate K area error predictors; And analyzing the area error prediction complexity according to the reflective image features to obtain error prediction complexity, calculating the error prediction of the wound area based on the error prediction complexity, K area error predictors and the reflective image features, and outputting a predicted area error.

Description

Machine vision-assisted diabetes wound surface image noise suppression method and system Technical Field The application relates to the technical field of medical imaging, in particular to a machine vision-assisted diabetes wound surface image noise suppression method and system. Background Along with the gradual rise of the incidence rate of diabetes, the diabetic foot has the characteristics of high disability rate and mortality rate, and forms serious threat to the life quality and life health of patients. In an actual clinical environment, due to the influence of factors such as shooting conditions, characteristics of a wound surface and body positions of a patient, various noise interferences often exist in acquired diabetes wound surface images, wherein reflection noise is one of the most common interference types. The reflective noise not only can mask the real structural information of the wound surface to cause the boundary blurring between the wound surface area and normal tissues in the image, but also can interfere the subsequent processing procedures of image segmentation, feature extraction, area calculation and the like. However, in the prior art, most of image noise suppression methods are universal denoising means, the specificity of reflection noise in diabetes wound images and the relevance of the reflection noise to wound tissue characteristics are not fully considered, loss or excessive smoothness of wound detail information is easily caused while the reflection noise is suppressed, and the calculation accuracy of wound area is insufficient, so that the denoising effect is not ideal. Disclosure of Invention The embodiment of the application solves the technical problem that the accuracy of wound area calculation is insufficient due to the fact that the interference caused by reflection of a wound area cannot be effectively solved by the traditional diabetes foot wound image noise suppression method by providing the machine vision-assisted diabetes wound image noise suppression method and system. The technical scheme for solving the technical problems is as follows: In a first aspect, the present application provides a machine vision-assisted diabetes wound image noise suppression method, the method comprising: extracting the reflective image characteristics of the reflective area of the diabetic foot wound surface image through a convolutional neural network; Determining a reflection interference type according to the reflection image feature analysis; If the reflection interference type is exudate reflection, predicting a wound area calculation error based on the reflection image characteristics to obtain a predicted area error; And formulating an adaptive image denoising mechanism according to the predicted area error, and performing reflection noise inhibition treatment on the diabetic foot wound surface image. In a second aspect, the present application provides a machine vision-assisted diabetes wound image noise suppression system comprising: the image feature extraction module is used for extracting the reflective image features of the reflective area of the diabetic foot wound surface image through the convolutional neural network; The interference type judging module is used for determining the reflection interference type according to the reflection image characteristic analysis; The wound surface error prediction module is used for predicting the calculation error of the wound surface area based on the reflection image characteristics to obtain the prediction area error if the reflection interference type is the exudate reflection; and the wound surface image processing module is used for formulating an adaptive image denoising mechanism according to the prediction area error and carrying out reflection noise inhibition processing on the diabetic foot wound surface image. The application provides one or more technical schemes, which at least have the following technical effects or advantages: The embodiment of the application provides a machine vision-assisted diabetes wound surface image noise suppression method and system, and firstly, a convolutional neural network is used for extracting the reflection image characteristics of a reflection area in a diabetes foot wound surface image. And secondly, determining the type of the reflective interference according to a preset characteristic identification rule table, and adopting a differentiated processing strategy to ensure that the reflective interference of different types is targeted inhibited. And thirdly, for the exudates reflective type, predicting the wound surface area calculation error based on reflective image characteristics, and matching an adaptive image denoising mechanism from an area error-denoising strategy mapping table according to the prediction error, so as to realize the accurate suppression of reflective noise. And finally, inquiring a pre-constructed exudates reflective area error-denoising strategy map