CN-121147701-B - Predictive model training method for obtaining blood perfusion map of keloid

CN121147701BCN 121147701 BCN121147701 BCN 121147701BCN-121147701-B

Abstract

The invention relates to a prediction model training method for acquiring blood perfusion pictures of keloids, which comprises the steps of acquiring a sample data set, preprocessing the sample data set, acquiring a training data set comprising a training set and a testing set, wherein each sample is a pix2pix sample and comprises a true general image, a true blood perfusion picture to which the true general image belongs, a true segmentation image of the general image, respectively inputting three images of each sample into four corresponding generators in a pre-constructed TRIPLEGAN neural network, and acquiring total loss information by three discriminators based on the output of the four generators and the corresponding images in the samples, further training and testing the TRIPLEGAN neural network to obtain a trained TRIPLEGAN neural network, and receiving the general image in the TRIPLEGAN neural network and outputting the generator of the blood perfusion picture as a prediction model after training. The method solves the problems of inaccurate evaluation of the traditional medium scale, high cost, time and labor waste caused by adopting LSCI for evaluation.

Inventors

YU NANZE
LONG XIAO
WANG XIAOJUN
LI SHUO
ZHANG MENGDI

Assignees

中国医学科学院北京协和医院

Dates

Publication Date: 20260508
Application Date: 20250904

Claims (8)

1. A predictive model training method for obtaining a blood perfusion map of a keloid, comprising: Obtaining a sample data set, wherein each sample of the sample data set comprises a general image of at least one keloid, a blood perfusion map corresponding to the general image, a general image outlining the boundary of the keloid, a blood perfusion map corresponding to the general image, and a blood perfusion map corresponding to the blood perfusion map; Preprocessing the sample data set to obtain a training data set comprising a training set and a testing set, wherein the training set comprises paired samples, and each paired sample is a pix2pix sample and comprises a true general image, a true segmentation image of the general image and a true blood flow perfusion image of the true general image; Respectively inputting the three images of each sample into four corresponding generators in a TRIPLEGAN neural network constructed in advance, and acquiring total loss information by three discriminators of the TRIPLEGAN neural network based on the output of the four generators and the corresponding images in the samples; the three images of each sample are respectively input into four corresponding generators in a TRIPLEGAN neural network constructed in advance, and the method comprises the following steps: For each paired sample: The true general image is input to a first generator G I2P , and a first blood flow perfusion map is output; The true blood perfusion map is input to the second generator G P2I , which outputs a first general image; The true general image is input to the third generator G I2S , and the first divided image is output; The true segmentation image is input to a G S2I of a fourth generator, and a second general image is output; The first perfusion map is input to the first generator G I2P , outputting a third general image; the first general image is input to a second generator G P2I , and a second blood flow perfusion map is output; the first divided image is input to the third generator G I2S , and a fourth general image is output; the second general image is input to the G S2I of the fourth generator, and a second divided image is output; Training and testing the TRIPLEGAN neural network based on the training set, the testing set and the preset target value of the total loss information to obtain a TRIPLEGAN neural network after training, and taking a generator which receives the general image in the TRIPLEGAN neural network and outputs a blood flow perfusion map as a prediction model after training.
2. The training method of claim 1, wherein the three discriminators of the TRIPLEGAN neural network obtain total loss information based on the outputs of the four generators and the corresponding images in the sample, comprising: After each sample is input to a generator, respectively acquiring a discriminator loss, a consistency loss and a circulation loss; The total loss information L comprises a discriminator loss L D which belongs to four generators, a circulation loss L cyc which belongs to four generators and two consistency losses L cons .
3. The training method of claim 1, wherein preprocessing the sample data set to obtain a training data set comprising a training set and a test set comprises: acquiring historical diagnosis and treatment information of a patient ID to which a general image belongs; Removing a general image of keloid treatment record information, epidermis incomplete information and a blood flow perfusion map of the general image existing in the sample data set based on the historical diagnosis and treatment information, and Removing a general image with appointed defect information in historical diagnosis and treatment information in a sample data set and a blood flow perfusion map of the general image; Cutting all the remaining general images and blood perfusion images and adopting a boundary registration mode to obtain a true general image of pix2pix and a true blood perfusion image to which the true general image belongs; And automatically identifying and processing the general image outlined with the keloid boundary to obtain a binarized true segmentation image; the training set comprises paired samples, and the test set comprises paired samples and unpaired samples; the true gross image and true blood perfusion map in the unpaired sample are not paired.
4. The training method of claim 1, wherein, The pre-constructed TRIPLEGAN neural network comprises four generators and three discriminators, wherein each generator is an antagonistic neural network, training and testing are carried out on the TRIPLEGAN neural network based on a training set, a testing set and a target value of preset total loss information to obtain a TRIPLEGAN neural network after training, The method comprises the steps of adopting a staged training method, firstly inputting a non-paired sample for training to serve as a first stage, then inputting a paired sample for training to serve as a second stage, and alternately carrying out the first stage and the second stage, wherein each cycle is more than 200 rounds; none of the losses in unpaired sample training is a resultant loss.
5. The training method of claim 1, wherein, The blood flow perfusion map of each paired sample is obtained by LSCI on the general image in the sample, and the blood flow perfusion value corresponding to the blood flow perfusion map is obtained by LSCI; Training the predictive model, further comprises: For a paired sample, obtaining a blood flow perfusion value calculated by a prediction model based on an output blood flow perfusion map, and comparing the blood flow perfusion value with a blood flow perfusion value in the paired sample; the trained predictive model is capable of outputting a blood flow perfusion map and blood flow perfusion values.
6. A method of predicting blood perfusion outcome of a keloid, comprising: acquiring a general image of a keloid of a specified size having a size to be analyzed; inputting the general image into a trained generator, and obtaining a blood flow perfusion result output by the generator; the trained generator is a generator obtained based on the training method for obtaining a predictive model of a blood perfusion map of keloids according to any of claims 1 to 5.
7. The method of claim 6, wherein inputting the general image into a trained generator to obtain a perfusion result of the blood flow output by the generator, comprising: If the prediction interface receives an instruction for predicting the blood flow perfusion map triggered by a user, inputting the general image into a trained generator, acquiring the blood flow perfusion map output by the generator, and comparing and displaying the general image and the predicted blood flow perfusion map at the prediction interface; If the prediction interface receives an instruction for predicting a blood flow perfusion value triggered by a user, inputting the general image into a trained generator, obtaining a blood flow perfusion chart output by the generator, and automatically generating a blood flow perfusion value based on the blood flow perfusion chart, wherein the blood flow perfusion value is an average value of blood flow pixels in a boundary range of keloids; And displaying the general image, the predicted blood flow perfusion map and the blood flow perfusion value at a prediction interface, wherein the blood flow perfusion map displays the boundary of the keloid.
8. An electronic device comprising a memory and a processor, the memory having stored therein a computer program, the processor executing the computer program stored in the memory and executing the steps of the method for predicting the blood perfusion result of a keloid according to any one of claims 6 and 7, or executing the steps of the training method for obtaining a prediction model of the blood perfusion map of a keloid according to any one of claims 1 to 5.

Description

Predictive model training method for obtaining blood perfusion map of keloid Technical Field The invention relates to the technical field of computers, in particular to a prediction model training method for acquiring a blood flow perfusion map of keloids, a prediction method for blood flow perfusion results of keloids and electronic equipment. Background In the keloid diagnosis and treatment process, the severity of the progress of the keloid is accurately evaluated, and the keloid diagnosis and treatment method is very necessary for reminding a patient to seek medical intervention, monitoring curative effect and objectively comparing curative effects after different intervention measures. At present, most of research on keloid evaluation at home and abroad mainly adopts scale evaluation, and the most common evaluation standard is the Vancouver Scar Scale (VSS). They are mainly subjectively evaluated in terms of pigmentation, vasodilation, thickness, flexibility, roughness, etc., VSS presence relies on observer subjectivity, and the disadvantages are poor reproducibility of observations and poor reproducibility between observers. Blood perfusion is an objective indicator of keloid assessment, but previous studies have not been consistent with blood flow assessment of keloid interiors. The reason may be that most research at present focuses on the number of blood vessels, which may not accurately reflect the state of blood perfusion inside pathological keloids, because blood vessels inside pathological scars (including hypertrophic scars and keloids) may appear as vascular endothelial hyperplasia, luminal stenosis, or even occlusion. Therefore, measuring and studying blood perfusion inside pathological keloids as a whole is a current hotspot. Laser speckle imaging (LSCI) is a recently more established technique based on speckle contrast analysis, and an innovative method for assessing tissue blood perfusion with high resolution and fast scan time provides objective facts and non-contact measurements of blood perfusion in specific areas. LSCI has been used to assess blood flow in microvascular disease, wine stain and burn keloid patients, with particular advantages including high image resolution, fast imaging speed, large scan range, low spatial variability, etc. LSCI is currently being used in a small number of studies to explore the level of blood perfusion in the field of keloids. Although LSCI has become an objective method of assessing blood flow perfusion of tissue, many hospitals do not have such LSCI blood flow assessment instruments on the one hand, and more time is still required to assess keloids using LSCI. Among them, the keloid is most time-consuming to outline, especially for multiple irregular keloids, which is much longer than single-shot keloid with regular morphology, which brings inconvenience to clinical work. For this reason, a solution for achieving automatic objective prediction of blood perfusion results of keloids using a general image is needed. Disclosure of Invention First, the technical problem to be solved In view of the above-mentioned shortcomings and disadvantages of the prior art, the present invention provides a training method for a prediction model for obtaining a blood perfusion map of keloids, a method for predicting a blood perfusion result of keloids, and an electronic device, which solve the problems of inaccurate scale evaluation and high cost and time and effort consumption caused by adopting LSCI for evaluation in the prior art. (II) technical scheme In order to achieve the above purpose, the main technical scheme adopted by the invention comprises the following steps: In a first aspect, an embodiment of the present invention provides a method for training a predictive model for obtaining a blood perfusion map of a keloid, comprising: Obtaining a sample data set, wherein each sample of the sample data set comprises a general image of at least one keloid, a blood perfusion map corresponding to the general image, a general image outlining the boundary of the keloid, a blood perfusion map corresponding to the general image, and a blood perfusion map corresponding to the blood perfusion map; Preprocessing the sample data set to obtain a training data set comprising a training set and a testing set, wherein the training set comprises paired samples, and each paired sample is a pix2pix sample and comprises a true general image, a true segmentation image of the general image and a true blood flow perfusion image of the true general image; Respectively inputting the three images of each sample into four corresponding generators in a TRIPLEGAN neural network constructed in advance, and acquiring total loss information by three discriminators of the TRIPLEGAN neural network based on the output of the four generators and the corresponding images in the samples; Training and testing the TRIPLEGAN neural network based on the training set, the testing set and the preset targe