A method and system for assessing liver fibrosis

Abstract

The present application relates to a kind of liver fibrosis evaluation method and system, belong to image processing technical field, the method includes: the target plasma of the person to be measured is obtained plasma dry film;Image acquisition is carried out to plasma dry film, obtain high-resolution image, and the target image is obtained by pre-processing high-resolution image;Based on target clinical task, target image is input into the deep learning model corresponding to target clinical task, obtain liver fibrosis evaluation result, wherein, target clinical task includes auxiliary distinction healthy and liver fibrosis, auxiliary screening liver fibrosis risk or auxiliary identification early liver fibrosis.The method can simultaneously support auxiliary distinction healthy and liver fibrosis, auxiliary screening liver fibrosis risk and auxiliary identification early liver fibrosis, improve the accuracy and efficiency of liver fibrosis evaluation, avoid the trauma and risk of puncture biopsy.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method and system for assessing liver fibrosis. Background Technology

[0002] Liver fibrosis is a critical pathological stage in the progression of chronic liver disease to cirrhosis and hepatocellular carcinoma. Early identification and accurate assessment are crucial for improving patient prognosis. Currently, commonly used clinical assessment methods include liver biopsy, serological markers, and elastography. However, liver biopsy is invasive and carries risks of sampling error and observer variability; serological markers are easily affected by factors such as inflammation, resulting in insufficient accuracy; and elastography relies on specialized equipment, is costly, and has limited applicability. Therefore, how to better assess liver fibrosis has become an urgent problem to be solved. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the purpose of the invention is to provide a method and system for assessing liver fibrosis.

[0004] In a first aspect, the present invention provides a method for assessing liver fibrosis, comprising: acquiring a target plasma smear from a subject; acquiring an image of the plasma smear to obtain a high-resolution image, and preprocessing the high-resolution image to obtain a target image; and inputting the target image into a deep learning model corresponding to a target clinical task to obtain a liver fibrosis assessment result, wherein the target clinical task includes assisting in distinguishing between health and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis.

[0005] A second aspect of the present invention provides a liver fibrosis assessment system, comprising: an acquisition module for acquiring a target plasma smear from a subject; an acquisition module for acquiring images of the plasma smear to obtain a high-resolution image, and preprocessing the high-resolution image to obtain a target image; and an acquisition module for inputting the target image into a deep learning model corresponding to a target clinical task to obtain a liver fibrosis assessment result, wherein the target clinical task includes assisting in distinguishing between health and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis.

[0006] A third aspect of the present invention provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method described in any one aspect of the present invention.

[0007] A fourth aspect of the present invention provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method described in any one of the first aspects of the present invention.

[0008] The beneficial effects of this invention are as follows: The liver fibrosis assessment method and system of this invention involves acquiring dried plasma smears of the target plasma from the subject; acquiring high-resolution images of the dried plasma smears; preprocessing the high-resolution images to obtain the target image; and inputting the target image into a deep learning model corresponding to the target clinical task to obtain the liver fibrosis assessment result. The target clinical task includes assisting in distinguishing between healthy individuals and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early-stage liver fibrosis. This method can simultaneously support the auxiliary distinction between healthy individuals and liver fibrosis, the auxiliary screening for liver fibrosis risk, and the auxiliary identification of early-stage liver fibrosis, improving the accuracy and efficiency of liver fibrosis assessment and avoiding the trauma and risks of puncture biopsy. Attached Figure Description

[0009] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. It is obvious that the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings.

[0010] Figure 1 This is a flowchart of a liver fibrosis assessment method according to an embodiment of the present invention; Figure 2 This is a flowchart of a liver fibrosis assessment method according to a specific embodiment of the present invention; Figure 3 This is a schematic diagram of a liver fibrosis assessment system according to an embodiment of the present invention; Figure 4 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0012] Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts disclosed in this invention.

[0013] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The terms "installed," "connected," and "linked" should be interpreted broadly; for example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of methods and systems consistent with some aspects of the invention as detailed in the appended claims.

[0015] This invention proposes a method, system, and related equipment for assessing liver fibrosis. Specifically, the following description of the method, system, and related equipment for assessing liver fibrosis according to embodiments of this invention is based on the accompanying drawings.

[0016] Figure 1 This is a flowchart of a liver fibrosis assessment method according to an embodiment of the present invention. It should be noted that the liver fibrosis assessment method of this embodiment can be applied to the liver fibrosis assessment system of this embodiment. This liver fibrosis assessment system can be configured on an electronic device or in a server. This application does not limit the scope of the application.

[0017] like Figure 1 As shown, the methods for assessing liver fibrosis include: S110, Obtain a dry plasma smear of the target plasma from the subject.

[0018] In an embodiment of the present invention, peripheral blood of the subject is collected, and plasma is separated. A precise pipette is used to quantitatively aspirate 10 microliters of plasma and drop it onto the center of a detection substrate. The detection substrate containing the plasma droplet is placed under constant temperature conditions and dried at 37°C–40°C for 10–20 minutes to completely solidify the droplet and form a dry plasma film with reproducible morphological characteristics. This allows for the acquisition of a dry plasma film of the target plasma from the subject.

[0019] The procedure involves collecting 2-5 mL of peripheral venous blood from the subject using an anticoagulant tube containing ethylenediaminetetraacetic acid (EDTA). After collection, the blood is centrifuged at 3500 rpm (×g) for 5 minutes within a preset time to separate the supernatant plasma. The plasma is transferred to a 1.5 mL EP tube. If immediate testing is not possible, the plasma can be frozen at -80°C for later use. Before testing, the frozen plasma should be thawed at room temperature.

[0020] Clean, adhesive glass slides were selected as the detection substrates. Before use, the slide surface was cleaned with distilled water and dried with lint-free paper, then preheated to 37°C in a constant temperature oven.

[0021] Using a precision pipette (2-20 μL range), aspirate 10 μL of thawed plasma. Place the droplet vertically and smoothly onto the center of the preheated test substrate. During addition, the pipette tip should be as close to the substrate surface as possible but not in contact with it to avoid air bubbles.

[0022] The substrate containing 10 μL of plasma droplets is transferred to a constant temperature and humidity chamber set at 37℃~40℃. It is then dried for 15 minutes at 37℃~40℃. To ensure complete curing of the pattern, a drying time range of 10-20 minutes can be selected depending on the ambient humidity, until the droplets are completely cured, forming a stable, circular solid sheet with a unique ring and crack pattern, i.e., a dried plasma sheet.

[0023] S120: Acquire images of dried plasma smears to obtain high-resolution images, and preprocess the high-resolution images to obtain the target image.

[0024] In embodiments of the present invention, when dried plasma slides are obtained, images can be acquired from the dried plasma slides to obtain high-resolution images. For example, the solidified dried plasma slides can be placed on the stage of a high-resolution slide scanner or microscope, and images can be acquired under preset parameters to obtain high-resolution images. The high-resolution images are color panoramic images.

[0025] The preset parameters include uniform resolution (e.g., ≥600dpi), color mode (24-bit color), brightness, and contrast.

[0026] In this process, when a high-resolution image is obtained, the image is subjected to size unification, pixel value normalization, and optional image enhancement operations to obtain a target image that meets the input requirements of the deep learning model.

[0027] S130: Based on the target clinical task, the target image is input into the deep learning model corresponding to the target clinical task to obtain the liver fibrosis assessment result.

[0028] By standardizing the acquisition and modeling analysis of droplet-dried images formed from samples of different liver states, intelligent identification and hierarchical assessment of liver fibrosis-related states can be achieved.

[0029] Among the target clinical tasks are to help distinguish between healthy individuals and liver fibrosis, to help screen for liver fibrosis risk, or to help identify early-stage liver fibrosis.

[0030] In other words, based on different target clinical tasks, the target image is input into the deep learning model corresponding to the target clinical task, thereby obtaining the liver fibrosis assessment result. Specific implementation methods can be found in subsequent embodiments.

[0031] According to the liver fibrosis assessment method of this invention, a dried plasma smear of the target plasma from the subject is obtained; the dried plasma smear is image-acquired to obtain a high-resolution image, and the high-resolution image is preprocessed to obtain a target image; based on a target clinical task, the target image is input into a deep learning model corresponding to the target clinical task to obtain the liver fibrosis assessment result. The target clinical task includes assisting in distinguishing between healthy individuals and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis. This method can simultaneously support the auxiliary distinction between healthy individuals and liver fibrosis, the auxiliary screening for liver fibrosis risk, and the auxiliary identification of early liver fibrosis, improving the accuracy and efficiency of liver fibrosis assessment and avoiding the trauma and risks of puncture biopsy.

[0032] To enable those skilled in the art to more readily understand the present invention, Figure 2 This is a liver fibrosis assessment method according to a specific embodiment of the present invention, such as... Figure 2 As shown, this method for assessing liver fibrosis includes: S210, Obtain a dry plasma slab of the target plasma from the subject; S220 acquires images of dried plasma smears to obtain high-resolution images, and preprocesses the high-resolution images to obtain the target image.

[0033] In the embodiments of the present invention, the implementation of steps S210-S220 can refer to the implementation of steps S110-S110 described above, and the present invention will not repeat the details.

[0034] S230, Define the target clinical task.

[0035] In embodiments of the present invention, the target clinical tasks include assisting in distinguishing between health and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis.

[0036] S240, based on the target clinical task, inputs the target image into the deep learning model corresponding to the target clinical task to obtain the assessment results related to liver fibrosis.

[0037] In embodiments of the present invention, when the target clinical task is to assist in distinguishing between health and liver fibrosis, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as healthy or liver fibrosis; when the target clinical task is to assist in screening for liver fibrosis risk, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as no fibrosis or having different degrees of fibrosis; when the target clinical task is to assist in identifying early liver fibrosis, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as no fibrosis or early fibrosis.

[0038] In an embodiment of the present invention, the deep learning model corresponding to the target clinical task is trained in the following manner: An image sample set was obtained and standardized. The standardized image samples were then labeled, and a hierarchical grouping five-fold cross-validation method was used to divide the labeled image samples into training, testing, and validation sets. The labels were categorized as healthy or with liver fibrosis, non-fibrotic or with varying degrees of fibrosis, or non-fibrotic or early-stage fibrosis. The image samples in the training set were then input into a deep learning model built on a ConvNeXt deep convolutional neural network. The backbone network of the deep learning model based on the ConvNeXt deep convolutional neural network is ConvNeXt. V2 outputs the predicted probabilities of healthy or liver fibrosis, or the predicted probabilities of no fibrosis or fibrosis of varying degrees, or the predicted probabilities of no fibrosis or early fibrosis. It uses the cross-entropy loss function to calculate the difference between the predicted probabilities and the ground truth labels. The AdamW optimizer is used to update network parameters. A StepLR learning rate scheduling strategy and mixed-precision computing are used to accelerate training. Within each fold, training is performed on the training set and evaluated on the validation set. During training at each fold, the AUC metric of the validation set is monitored, and the model with the best AUC metric is automatically extracted from the cross-validation results of each fold as a candidate sub-model for that fold. Normalized fusion weights are assigned to the five candidate sub-models based on their AUC values. Image samples from the test set are input to the five candidate sub-models, each outputting a probability. The discrimination result is output based on the normalized fusion weights and the output probabilities of each candidate sub-model. After each training epoch, the performance metric is calculated on the validation set, and the optimal training epoch is selected based on the performance metric. The corresponding model weights are saved as the trained deep learning model.

[0039] In an embodiment of the present invention, when the target clinical task is to assist in distinguishing between health and liver fibrosis, the target image can be input into a deep learning model corresponding to the target clinical task, which outputs the probability of health or liver fibrosis; based on the probability of health or liver fibrosis corresponding to the target image, the evaluation result is determined. For example, if the probability of health is greater than a preset threshold, the evaluation result is health.

[0040] In an embodiment of the present invention, when the target clinical task is to assist in screening for the risk of liver fibrosis, the target image is input into the deep learning model corresponding to the target clinical task, and the model outputs the probability of no fibrosis or the presence of different degrees of fibrosis; based on the probability of no fibrosis or the presence of different degrees of fibrosis corresponding to the target image, the evaluation result is determined. For example, if the probability of no fibrosis is greater than a preset threshold, the evaluation result is no fibrosis.

[0041] In an embodiment of the present invention, when the target clinical task is to assist in the identification of early liver fibrosis, the target image is input into the deep learning model corresponding to the target clinical task, and the probability of no fibrosis or early fibrosis is output; based on the probability of fibrosis or early fibrosis corresponding to the target image, the evaluation result is determined. For example, if the probability of early fibrosis is greater than a preset threshold, the evaluation result is early fibrosis.

[0042] In the first embodiment of the present invention, a liver fibrosis screening model for healthy individuals based on AI and plasma droplet drying patterns was constructed and clinically validated. 1) Data preparation: The target images corresponding to individual plasma smears prepared through standardized procedures were divided into "healthy group" and "liver fibrosis group" according to clinical information, and then divided into training set and test set in an 8:2 ratio.

[0043] 2) Model Training: The training code is run using Python. First, a guidance file containing image paths, category labels, dataset partitioning annotations, and sample grouping identifiers is read, limiting modeling to the training set data. Then, stratification is performed based on category labels, and grouping is constrained based on sample_id. Five-fold cross-validation is implemented to prevent information leakage caused by data from the same sample source entering different subsets simultaneously. During model construction, the ConvNeXtV2 network is used as the feature extraction backbone, and pre-trained weights unrelated to the classification head are loaded for initialization to achieve transfer learning. Input images are uniformly sized, randomly flipped, color-perturbed, and standardized before being fed into the network for training. During training, cross-entropy loss, AdamW optimization algorithm, StepLR learning rate scheduling strategy, and mixed precision calculation mechanism are used to improve model parameter update efficiency and training stability.

[0044] 3) Model Construction: The model with the optimal validation set AUC is automatically extracted from the cross-validation results of each fold as a candidate sub-model, and normalized fusion weights are assigned based on the optimal AUC value of each sub-model. Then, the image to be tested is input into each candidate sub-model for inference to obtain the predicted probability of each category. Finally, the output probabilities of each sub-model are weighted and fused to obtain the final discrimination result. This method avoids the instability of single-model discrimination, integrates the discrimination ability of models trained with different data folds, and improves the robustness and generalization performance of the final model.

[0045] 4) Model Result Output: After the model training is completed, the accuracy, precision, recall, F1 score and AUC confusion matrix of each class are output in each round of validation. The model parameter file and evaluation result file of the corresponding round are saved at the same time.

[0046] 5) Real-world experimental results: This study compared the performance of four deep learning models in a health-liver fibrosis screening task. The results showed that all four models had high classification efficiency, with ConvNeXtV2 and Swin Transformer V2 performing best. ConvNeXtV2 achieved recognition rates of 100% and 99.8% for class 0 and class 1 samples, respectively, while Swin Transformer V2 achieved 99.7% and 100%, respectively, both close to ideal levels. In comparison, ResNet50 achieved 98.2% and 94%, respectively, and EfficientNet achieved 91.6% and 98.1%, respectively. Comprehensive metric analysis shows that ConvNeXtV2's AUC, Accuracy, Precision, Recall, and F1 score are 1.000, 1.000, 1.000, 1.000, and 1.000, respectively, while SwinTransformer V2's are 1.000, 0.997, 0.997, 0.996, and 0.997, respectively, outperforming ResNet50 and EfficientNet overall. ROC analysis further shows that the deep learning models' AUCs in the internal validation queue are all higher than the traditional metrics APRI (0.846) and FIB-4 (0.755); in the external validation queue, the final model's AUC reaches 0.985, also outperforming APRI's 0.862 and FIB-4's 0.741. Furthermore, the model's accuracy, precision, recall, and F1 score in the external queue were 0.934, 0.931, 0.923, and 0.949, respectively, indicating that the model still has good stability and generalization ability in independent external data.

[0047] In a second embodiment of the present invention, a screening model for high-risk populations of liver fibrosis based on AI and plasma droplet drying patterns is constructed and clinically validated. 1) Data preparation: The target images corresponding to the single plasma dry slices prepared through standardized procedures were divided into "liver disease but no liver fibrosis group (F0)" and "liver fibrosis group (F1-4)" according to clinical information, and were divided into training set and test set in an 8:2 ratio.

[0048] 2) Model Training: The training code is run using Python. The core process includes: constructing a sample set based on the image paths and binary labels preset in the guidance file; automatically removing invalid path samples; and using hierarchical five-fold cross-validation to divide the data to maintain the relative stability of the class distribution at each fold. In the model building stage, a pre-trained ConvNeXtV2 is used as the feature extraction backbone network, and the class weights are adaptively calculated based on the class distribution of the training set. These class weights are then introduced into the cross-entropy loss function to reduce the impact of class imbalance on model training. During training, the model is iteratively optimized using data augmentation methods such as random cropping, flipping, and color perturbation, as well as an automatic mixed precision calculation mechanism. The model is then validated in each round.

[0049] 3) Model construction: The model with the best AUC on the validation set is automatically extracted from the cross-validation results of each fold as a candidate sub-model, and normalized fusion weights are assigned according to the best AUC value of each sub-model. Then, the image to be tested is input into each candidate sub-model for inference to obtain the predicted probability of each category. The output probabilities of each sub-model are then weighted and fused to obtain the final discrimination result.

[0050] 4) Model Results Output: After model training, performance metrics such as accuracy, precision, recall, F1 score, AUC for each class, and confusion matrix are output at each fold and validation round. The model parameter files and evaluation result files for each round are saved simultaneously. The high-risk population screening model demonstrates high diagnostic efficacy in both the internal and external cohorts, with AUCs of 1.000 and 0.960, respectively, significantly outperforming the AUCs of traditional metrics APRI (internal: 0.804, external: 0.683) and FIB-4 (internal: 0.791, external: 0.683). Furthermore, the model's accuracy, F1 score, precision, and recall in the internal cohort are 1.000, 1.000, 1.000, and 1.000, respectively, while in the external cohort A, they are 0.897, 0.906, 0.964, and 0.854, respectively. Although the external validation performance declined slightly, it remained at a high level overall, indicating that the model has good screening value and external generalization ability.

[0051] In a third embodiment of the present invention, the construction and clinical validation of an early diagnostic model for liver fibrosis based on AI and plasma droplet drying patterns are described below. 1) Data preparation: The target images corresponding to the single plasma dry slices prepared through standardized procedures were divided into "liver disease but no liver fibrosis group (F0)" and "early liver fibrosis group (F1)" according to clinical information, and were divided into training set and test set in an 8:2 ratio.

[0052] 2) Model Training: The training code is run using Python. The core process includes: automatically reading images and building a dataset based on the sample paths and class label data defined in the Excel guide sheet; using a pre-trained ConvNeXtV2Base network as the feature extraction and classification backbone to complete binary classification of the input images; during training, preprocessing and data augmentation strategies such as random cropping, horizontal / vertical flipping, color perturbation, and standardization are employed, combined with a cross-entropy loss function with class weights, the AdamW optimizer, and automatic mixed precision to accelerate training, thereby improving the model's training stability and adaptability to imbalanced data; simultaneously, a five-fold hierarchical cross-validation mechanism is introduced, performing training, validation, round performance comparison, and model weight saving in each fold, and automatically selecting the optimal round for summarizing out-of-fold prediction results according to the rule of "AUC first, accuracy second best, validation loss third best"; 3) Model construction: While maintaining the same network structure parameter configuration as in the training phase, it calls the ConvNeXtV2 Base binary classification architecture to reconstruct the model framework and loads the pre-trained target weight file; then, according to the Excel guide, it reads the image path and label information of the samples to be verified, performs uniform size scaling, center cropping, tensor quantization and standardization preprocessing on the input images, and then sends them in batches into the model for forward inference, outputting the predicted probability of each image belonging to the positive category.

[0053] 4) Model Output Results: After model training, the output includes weight files for each round, validation prediction results for each round, round-by-round evaluation metrics, summary of optimal rounds for each fold, overall OOF prediction results, overall accuracy / precision / recall / F1 score / AUC / confusion matrix, and ROC curve data. The early liver fibrosis diagnostic model demonstrated high diagnostic efficacy in both the internal and external cohorts, with AUCs of 1.000 and 0.921, respectively, significantly outperforming the AUCs of traditional metrics APRI (internal: 0.652, external: 0.549) and FIB-4 (internal: 0.612, external: 0.545). Furthermore, the model's accuracy, F1 score, precision, and recall in the internal cohort were 1.000, 1.000, 1.000, and 1.000, respectively, while in external cohort A they were 0.901, 0.860, 0.908, and 0.817, respectively. Although the external validation performance declined slightly, it remained at a high level overall, indicating that the model has good screening value and external generalization ability.

[0054] In embodiments of the present invention, if the target image is obtained by acquiring images of dried plasma smears and preprocessing the acquired images, images of dried serum smears, dried blood spots, or dried urine smears can also be acquired. If the target image is obtained by preprocessing the acquired images, the target image is input into the deep learning model corresponding to the target clinical task based on the target clinical task to obtain the liver fibrosis assessment result.

[0055] In the embodiments of the present invention, in addition to ConvNeXt V2, the Swin Transformer V2 evaluated in the present invention, as well as more advanced CNN or ViT variants that may emerge in the future, can all be used as alternatives to the backbone network, as long as they can effectively capture the multi-scale morphological features of plasma smears.

[0056] In embodiments of the present invention, the features of the target image corresponding to the plasma smear are multimodally fused with the patient's clinical baseline data (such as age, gender, and medical history) or conventional serological indicators (such as ALT, AST, and APRI) to construct a hybrid diagnostic model that includes both image and non-image information, thereby further improving diagnostic performance. For example, the image feature vector extracted by ConvNeXt V2 is concatenated with clinical indicators and then input into a classifier for final decision-making.

[0057] In embodiments of the present invention, based on different target clinical tasks, the functions of the models corresponding to different target clinical tasks, i.e., three models, can be integrated into a single network with multiple output heads. By designing a joint loss function (such as weighted cross-entropy loss) for end-to-end training, a single model can output three diagnostic probabilities simultaneously, simplifying the deployment process and potentially improving efficiency.

[0058] The liver fibrosis assessment method according to embodiments of the present invention only requires peripheral blood collection to prepare plasma dried slides, avoiding the trauma and risks associated with traditional liver biopsy. It is simple to operate, highly reproducible, and suitable for large-scale population screening and dynamic monitoring. By strictly controlling the plasma volume (10 μL), drying temperature (37℃~40℃), time (10~20 minutes), and using an adhesive slide, stable and reproducible ring and crack patterns of plasma dried slides can be formed, providing consistent and reliable input images for subsequent deep learning models and solving the problem of high randomness in traditional droplet drying patterns. All three clinical tasks achieved near-perfect results in independent external validation cohorts. With an AUC exceeding 0.92, all metrics (accuracy, precision, recall, and F1 score) remain above 0.86, demonstrating the model's good adaptability and robustness to different populations and testing environments. The model employs five-fold cross-validation and sample ID-based grouping constraints to prevent samples from the same patient from appearing simultaneously in both the training and validation sets. A weighted fusion strategy (assigning normalized weights based on the AUC of each fold) integrates multiple candidate sub-models, resulting in greater stability and generalization performance compared to a single model. Introducing class weights into the loss function effectively addresses class imbalance issues (e.g., fewer early fibrosis samples). Transfer learning (ConvNeXt V2 pre-trained weights), data augmentation (random flipping, color perturbation), and mixed-precision training improve training efficiency and model performance. This method allows for the training of corresponding binary classification models for different clinical objectives (health screening, high-risk screening, early diagnosis), and clearly defines the label definitions and model output formats for each task, facilitating clinical selection as needed. The above scheme establishes a non-invasive, high-precision, and generalizable method for assessing liver fibrosis based on macroscopic morphological images of dried plasma slides and deep learning.

[0059] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0060] According to one aspect of the present invention, a liver fibrosis assessment system is also provided. Figure 3 This is a schematic diagram of a liver fibrosis assessment system according to an embodiment of the present invention; as shown. Figure 3 As shown, it includes: The acquisition module 310 is used to acquire plasma smears of the target plasma from the subject. The acquisition module 320 is used to acquire images of the dried blood plasma slices to obtain high-resolution images, and to preprocess the high-resolution images to obtain target images; The module 330 is used to input the target image into the deep learning model corresponding to the target clinical task based on the target clinical task to obtain the liver fibrosis assessment result. The target clinical task includes assisting in distinguishing between health and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis.

[0061] According to an embodiment of the present invention, a liver fibrosis assessment system acquires dried plasma smears of the target plasma from the subject; images of the dried plasma smears are acquired to obtain high-resolution images, and the high-resolution images are preprocessed to obtain target images; based on a target clinical task, the target images are input into a deep learning model corresponding to the target clinical task to obtain liver fibrosis assessment results. The target clinical task includes assisting in distinguishing between healthy individuals and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis. This system can simultaneously support the auxiliary distinction between healthy individuals and liver fibrosis, the auxiliary screening for liver fibrosis risk, and the auxiliary identification of early liver fibrosis, improving the accuracy and efficiency of liver fibrosis assessment and avoiding the trauma and risks of puncture biopsy.

[0062] Optionally, the method for preparing the target plasma dry film includes: taking peripheral blood from the subject and separating plasma; quantitatively aspirating 10 microliters of the plasma using a precision pipette and dropping it onto the center of a detection substrate; placing the detection substrate containing the plasma droplet under constant temperature conditions and drying it at 37°C to 40°C for 10 to 20 minutes to completely solidify the droplet and form a plasma dry film with repeatable morphological characteristics.

[0063] Optionally, the acquisition module 320 is specifically used to perform size unification, pixel value normalization, and optional image enhancement operations on the high-resolution image to obtain a target image that adapts to the input requirements of the deep learning model.

[0064] Optionally, when the target clinical task is to assist in distinguishing between healthy individuals and liver fibrosis, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as healthy or liver fibrosis; when the target clinical task is to assist in screening for liver fibrosis risk, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as no fibrosis or having different degrees of fibrosis; when the target clinical task is to assist in identifying early liver fibrosis, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as no fibrosis or early fibrosis.

[0065] Optionally, the deep learning model corresponding to the target clinical task is trained in the following manner: An image sample set is acquired, and the image samples in the image sample set are standardized; the standardized image samples are labeled, and the labeled image samples are divided into a training set, a test set, and a validation set using a hierarchical grouping five-fold cross-validation method, wherein the labels are either healthy or liver fibrosis, or non-fibrosis or fibrosis of varying degrees, or non-fibrosis or early fibrosis; the image samples in the training set are input into a deep learning model constructed based on a ConvNeXt deep convolutional neural network, wherein the backbone network of the deep learning model constructed based on the ConvNeXt deep convolutional neural network is ConvNeXt. V2 outputs the predicted probability of being healthy or having liver fibrosis, or the predicted probability of having no fibrosis or having fibrosis of varying degrees, or the predicted probability of having no fibrosis or early fibrosis; it calculates the difference between the predicted probability and the true labeled probability using the cross-entropy loss function; it updates the network parameters using the AdamW optimizer; it uses the StepLR learning rate scheduling strategy and enables mixed precision computation to accelerate training, wherein within each fold, training is performed on the training set and evaluation is performed on the validation set; during the training process of each fold, the AUC index of the validation set is monitored, and the predicted probability is automatically extracted from the cross-validation results of each fold. The model with the best AUC on the validation set is selected as a candidate sub-model for the next step. Based on the AUC values ​​of the five candidate sub-models, normalized fusion weights are assigned to each candidate sub-model. Image samples from the test set are input to the five candidate sub-models, each outputting a probability. A discrimination result is output based on the normalized fusion weights and the output probabilities of each candidate sub-model. After each training epoch, a performance metric is calculated on the validation set, and the optimal training epoch is selected based on the performance metric. The corresponding model weights are saved as the trained deep learning model.

[0066] Optionally, module 330 is configured to input the target image into a deep learning model corresponding to the target clinical task, outputting the probability of health or liver fibrosis; and determine the evaluation result based on the probability of health or liver fibrosis corresponding to the target image; or input the target image into a deep learning model corresponding to the target clinical task, outputting the probability of no fibrosis or the presence of different degrees of fibrosis; and determine the evaluation result based on the probability of no fibrosis or the presence of different degrees of fibrosis corresponding to the target image; or input the target image into a deep learning model corresponding to the target clinical task, outputting the probability of no fibrosis or early fibrosis; and determine the evaluation result based on the probability of no fibrosis or early fibrosis corresponding to the target image.

[0067] According to one aspect of the present invention, an electronic device is provided.

[0068] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Figure 4 As shown, an electronic device may include one or more ( Figure 4 Only one is shown in the image. A processor 102 (which may include, but is not limited to, a microprocessor unit (MPU) or a programmable logic device (PLD)) and a memory 104 for storing data are also shown. In one exemplary embodiment, the electronic device may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 4 The structure shown is for illustrative purposes only and does not limit the structure of the terminal device described above. For example, the terminal device may also include components that are more... Figure 4 The more or fewer components shown, or having the same Figure 4 Equivalent functions or ratios shown Figure 4 The functions shown have more different configurations.

[0069] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the liver fibrosis assessment method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to terminal devices via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0070] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the switching device. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0071] The present invention proposes a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform a liver fibrosis assessment method.

[0072] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

[0073] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0074] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing liver fibrosis, characterized in that, include: Obtain dried plasma slices of the target plasma from the subject; The plasma smear is imaged to obtain a high-resolution image, and the high-resolution image is preprocessed to obtain the target image; Based on the target clinical task, the target image is input into the deep learning model corresponding to the target clinical task to obtain the liver fibrosis assessment result. The target clinical task includes assisting in distinguishing between health and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis.

2. The liver fibrosis assessment method according to claim 1, characterized in that, The method for preparing the target plasma dried tablets includes: Peripheral blood was collected from the subject of the test, and plasma was separated. A precise pipette was used to quantitatively aspirate 10 microliters of the plasma and drop it onto the center of the detection substrate. The detection substrate containing plasma droplets is placed under constant temperature conditions and dried at 37°C to 40°C for 10 to 20 minutes to completely solidify the droplets and form a plasma dry film with repeatable morphological characteristics.

3. The liver fibrosis assessment method according to claim 1, characterized in that, The high-resolution image is preprocessed to obtain the target image, including: The high-resolution image is subjected to size unification, pixel value normalization, and optional image enhancement operations to obtain a target image that meets the input requirements of the deep learning model.

4. The liver fibrosis assessment method according to claim 1, characterized in that, When the target clinical task is to help distinguish between healthy and liver fibrosis, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as healthy or liver fibrosis. When the target clinical task is to assist in screening for liver fibrosis risk, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as either non-fibrotic or have different degrees of fibrosis. When the target clinical task is to assist in the identification of early liver fibrosis, the image samples in the training set of the deep learning model corresponding to the target clinical task are labeled as either non-fibrotic or early fibrotic.

5. The liver fibrosis assessment method according to claim 4, characterized in that, The deep learning model corresponding to the target clinical task is trained in the following way: Obtain an image sample set and perform standardization processing on the image samples in the image sample set; The standardized image samples are labeled, and the labeled image samples are divided into training set, test set and validation set using hierarchical grouping five-fold cross-validation. The labeled images are labeled as healthy or liver fibrosis, or as no fibrosis or fibrosis of different degrees, or as no fibrosis or early fibrosis. The image samples in the training set are input into a deep learning model built on a ConvNeXt deep convolutional neural network, wherein the backbone network of the deep learning model built on the ConvNeXt deep convolutional neural network is ConvNeXt V2, and the predicted probability of health or liver fibrosis is output, or the predicted probability of no fibrosis or the presence of different degrees of fibrosis is output, or the predicted probability of no fibrosis or early fibrosis is output. The difference between the predicted probability and the true label is calculated using the cross-entropy loss function; Update network parameters using the AdamW optimizer; We employ a StepLR learning rate scheduling strategy and enable mixed-precision computation to accelerate training, where within each iteration, we train on the training set and evaluate on the validation set. During the training process of each fold, the AUC index of the validation set is monitored, and the model with the best AUC index of the validation set is automatically extracted from the cross-validation results of each fold as a candidate sub-model of the fold. Based on the AUC values ​​corresponding to the five candidate sub-models, assign normalized fusion weights to the candidate sub-models; The image samples in the test set are input into the five candidate sub-models, each candidate sub-model outputs a probability, and the discrimination result is output based on the normalized fusion weights corresponding to the candidate sub-models and the output probability of each candidate sub-model; After each training epoch is completed, a performance metric is calculated on the validation set, and the optimal training epoch is selected based on the performance metric. The corresponding model weights are then saved as the trained deep learning model.

6. The liver fibrosis assessment method according to claim 5, characterized in that, The target image is input into the deep learning model corresponding to the target clinical task to obtain liver fibrosis assessment results, including: The target image is input into a deep learning model corresponding to the target clinical task, which outputs the probability of health or liver fibrosis; based on the probability of health or liver fibrosis corresponding to the target image, the evaluation result is determined; or The target image is input into the deep learning model corresponding to the target clinical task, and the probability of no fibrosis or the presence of different degrees of fibrosis is output; based on the probability of no fibrosis or the presence of different degrees of fibrosis corresponding to the target image, the evaluation result is determined; or The target image is input into the deep learning model corresponding to the target clinical task, and the probability of no fibrosis or early fibrosis is output; the evaluation result is determined based on the probability of no fibrosis or early fibrosis corresponding to the target image.

7. A liver fibrosis assessment system, characterized in that, include: The acquisition module is used to acquire dried plasma slices of the target plasma from the subject. The acquisition module is used to acquire images of the dried blood plasma smear to obtain high-resolution images, and to preprocess the high-resolution images to obtain target images; The module is used to input the target image into the deep learning model corresponding to the target clinical task based on the target clinical task to obtain the liver fibrosis assessment result. The target clinical task includes assisting in distinguishing between health and liver fibrosis, assisting in screening for liver fibrosis risk, or assisting in identifying early liver fibrosis.

8. The liver fibrosis assessment system according to claim 7, characterized in that, The acquisition module is specifically used for: The high-resolution image is subjected to size unification, pixel value normalization, and optional image enhancement operations to obtain a target image that meets the input requirements of the deep learning model.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 6.

Images