A method for predicting the progression of pulmonary fibrosis based on time-series CT images

Abstract

This invention discloses a method for predicting the progression of pulmonary fibrosis based on time-series CT images. By statistically analyzing all voxels within the entire lung region and all voxels within the fibrotic region, the percentage of fibrotic volume relative to the total lung volume is obtained. Based on time-series data showing the change in fibrosis proportion over time or the change in fibrosis proportion between the baseline time point and each follow-up time point, the annual average fibrosis growth rate is calculated. Based on the time-series data showing the change in fibrosis proportion over time, a time-fibrosis proportion mapping model is constructed. Based on the annual average fibrosis growth rate, a progression type map is plotted, and survival curves for various progression types and progression models corresponding to different progression types are established. Based on the above, a pulmonary fibrosis progression prediction model based on time-series CT images is constructed. This model can define an index of fibrosis growth rate and can automatically calculate the fibrosis growth rate using artificial intelligence algorithms, and use it to predict the clinical prognosis of fibrotic diseases such as IPF.

Description

Technical Field

[0001] This invention belongs to the field of medical artificial intelligence technology, specifically relating to a method for predicting the progression of pulmonary fibrosis based on time-series CT images, which is used to achieve quantitative prediction of pulmonary fibrosis progression, group assessment, and correlation with survival prognosis. Background Technology

[0002] Idiopathic pulmonary fibrosis (IPF) is a disease of unknown cause characterized by diffuse alveolitis and alveolar structural disorder, ultimately leading to interstitial fibrosis of the lungs. IPF is progressive; most patients experience relatively slow progression over time, while a minority experience rapid decline and death within months. Furthermore, the clinical course of IPF can be interrupted by acute respiratory exacerbations (AEs), which further complicate the course of IPF.

[0003] Traditional manual visual assessment is highly subjective and prone to error, failing to accurately obtain fibrosis growth rate data. Existing quantitative CT methods for IPF (such as histogram kurtosis method, CALIPER method, data-driven texture analysis, and quantitative methods for pulmonary fibrosis) can only reflect the fibrosis level at a single time point, failing to accurately define and obtain the fibrosis growth rate indicator. Furthermore, no correlation model has been established between the fibrosis growth rate and patient survival prognosis, making it impossible to predict future progression. Other models in the field rely on multiple clinical indicators, which are difficult to obtain and thus cannot meet the needs of practical clinical applications.

[0004] While time-series CT has made breakthroughs in predicting disease progression in lung cancer, abdominal aortic aneurysm, and colorectal cancer lung metastases, it has not yet been effectively applied to interstitial lung disease (IPF). Therefore, there is an urgent need for a standardized and quantitative prediction method based on time-series CT images, combined with fibrosis rate of increase grouping and survival association analysis, to achieve dynamic prediction and prognostic assessment of IPF progression, and to address the problems of insufficient clinical understanding of disease progression and lack of reliable basis for treatment decisions. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention provides a method for predicting the progression of pulmonary fibrosis based on time-series CT images, comprising the following steps:

[0006] Baseline, follow-up CT images, and survival follow-up information of multicenter IPF patients were collected to construct a time-series sequence of CT images; The time-series CT images were preprocessed, followed by segmentation of the whole lung region and the fibrotic region. All voxels in the whole lung region and all voxels in the fibrotic region were counted. The whole lung volume and fibrotic volume were calculated by combining the corresponding voxel spatial resolution and normalized to obtain the percentage of fibrotic volume to the whole lung volume at different time points, forming time-series data of fibrosis ratio changing over time. The annual average fibrosis growth rate was calculated based on the time-series data of fibrosis ratio changing over time or the change in fibrosis ratio between the baseline time point and each follow-up time point. Based on time-series data of fibrosis proportion changing over time, a time-fibrosis proportion mapping model was constructed. A progression type map was plotted based on the annual average fibrosis growth rate. The KM survival analysis method was used to establish survival curves for multiple progression types. Corresponding progression models were constructed for patients with different progression types, so that patients in the same group could share similar progression pattern parameters. Based on the baseline, follow-up CT images, survival follow-up information, percentage of fibrosis volume to total lung volume, time-series data of fibrosis proportion changing over time, annual average fibrosis growth rate, progression type map, survival curves for multiple progression types, and progression models corresponding to different progression types, a pulmonary fibrosis progression prediction model based on time-series CT images was constructed for multicenter IPF patients. The patient's baseline, follow-up CT images, and survival follow-up information are input into a pulmonary fibrosis progression prediction model based on time-series CT images. The model outputs the percentage of predicted fibrosis volume in the total lung volume, the progression group, and the corresponding survival prognosis.

[0007] Preferably, the survival follow-up information includes survival status and follow-up cutoff time.

[0008] Preferably, the preprocessing includes nonlinear denoising and image alignment.

[0009] Preferably, the time-fiber ratio mapping model includes: A parametric trend model is used to describe the continuous change process of the fiber ratio. A piecewise modeling progression function is used to characterize the changes in the rate of fibrosis progression at different time stages; Nonlinear growth models are used to describe the acceleration or deceleration characteristics of fibrosis at different stages of the disease. Time series models based on probability or statistical inference are used to characterize the uncertainty of fibrosis progression.

[0010] Preferably, the parameterized trend model adopts a simplified trend function form, which uses a small number of parameters to characterize the patient's baseline fibrosis burden and its rate of progression over time. It is obtained by fitting or estimating follow-up data at multiple time points, so that it can reflect the fibrosis progression characteristics at the individual patient level.

[0011] This invention provides a method for predicting the progression of pulmonary fibrosis based on time-series CT images. By statistically analyzing all voxels within the entire lung region and all voxels within the fibrotic region, the percentage of fibrotic volume relative to the total lung volume is obtained. The annual average fibrosis growth rate is calculated based on time-series data showing the change in fibrosis proportion over time, or the change in fibrosis proportion between the baseline time point and each follow-up time point. A time-fibrosis proportion mapping model is constructed based on the time-series data showing the change in fibrosis proportion over time. A progression type map is plotted based on the annual average fibrosis growth rate. Survival curves for various progression types and progression models corresponding to different progression types are established. Based on the above, a pulmonary fibrosis progression prediction model based on time-series CT images is constructed. This model can define an index for the fibrosis growth rate and automatically calculate the fibrosis growth rate using artificial intelligence algorithms, and use it to predict the clinical prognosis of fibrotic diseases such as IPF. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the CT image time sequence provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the advancements in the embodiments of the present invention; Figure 3 Schematic diagrams of survival curves and progression models for various progression types provided in embodiments of the present invention; Figure 4 This is a schematic diagram of the output data of a pulmonary fibrosis progression prediction model based on time-series CT images provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0014] This invention provides a method for predicting the progression of pulmonary fibrosis based on time-series CT images, comprising the following steps: Baseline, follow-up CT images, and survival follow-up information of IPF patients from multiple centers were collected, and the specific time of each CT examination was recorded simultaneously. The follow-up period was no less than 36 months to ensure that the average annual fibrosis growth rate and long-term survival data could be calculated.

[0015] For patients with CT scans more than once every 6 months, images were selected at 6-month intervals to construct a CT image time sequence, such as... Figure 1 As shown.

[0016] For examinations conducted less than every 6 months, all actual examination images should be retained to construct a time-series CT image sequence, ensuring the continuity and integrity of the time-series data.

[0017] Survival follow-up information includes survival status, follow-up cutoff time, etc., which is used for subsequent survival prognostic association analysis.

[0018] Denoising: The original CT sequence was resampled to an isotropic spatial resolution of 1.0mm×1.0mm×1.0mm using trilinear interpolation to ensure that the voxel volume r remains constant in all time-series samples and to eliminate interference from slice thickness differences. Subsequently, a median filter was used for nonlinear denoising to reduce statistical noise caused by the random nature of X-rays and improve image quality.

[0019] Dynamic thresholding of CT values: Based on lung tissue characteristics, the original Hounsfield Unit (HU) values ​​are restricted to the range of [-1000, 400] and mapped to the [0, 1] space using a linear normalization function, aiming to enhance the contrast between fibrotic areas (such as honeycomb shadows and reticular shadows) and normal lung tissue.

[0020] Multi-scale temporal spatial registration and image alignment: High-quality and structurally significant CT images are selected as reference images in the temporal sequence of CT images. Feature markers of images at each time point are extracted. Transformation parameters such as translation, rotation, and scaling are estimated by matching to ensure that all temporal images are spatially consistent.

[0021] Automatic segmentation model for region segmentation: Based on the U-net deep learning model, regions of interest (ROIs) are extracted from aligned CT images, and the whole lung region and fibrotic region are segmented. The automatic segmentation model is trained using a dataset annotated by professional physicians (training set:test set = 8:2), and the segmentation accuracy is verified based on the annotation results to ensure the reliability of the extraction of the whole lung region and fibrotic region.

[0022] (1) Calculation of the fiber ratio For a CT image acquired at any given time point, the whole lung region and fibrotic region at that time point are first obtained based on an automatic segmentation model.

[0023] The total lung volume is calculated by statistically analyzing all voxels within the entire lung region and combining them with the corresponding voxel spatial resolution, using the following formula:

[0024] In the formula, for Total lung volume at time points for The total number of voxels in the entire lung region at a given time point. Voxel spatial resolution (unit: mm³ / voxel) of CT images is determined by the scanning parameters of the CT equipment.

[0025] The fibrosis volume is calculated by statistically analyzing voxels within the fibrosis region and combining this with spatial resolution, using the following formula:

[0026] In the formula, for Volume of the fibrotic region at a given time point for The total number of voxels in the fibrotic region at a given time point.

[0027] Based on this, the fibrotic volume and the total lung volume were normalized to obtain the percentage of fibrotic volume to the total lung volume at that time point. The fibrosis percentage and degree of each lung segment were calculated in the same way, forming time-series data on the change of fibrosis ratio over time. This data characterizes the overall degree of pulmonary fibrosis in the patient at the time of the CT scan. This fibrosis ratio, as a standardized indicator, can eliminate the influence of differences in lung volume among different patients and different scanning parameters on the absolute volume calculation, thus exhibiting good comparability.

[0028] The formula for the percentage of fibrous volume is as follows: % In the formula, for The percentage of pulmonary fibrosis volume to total lung volume at a given time point is the core standardized indicator of this invention, characterizing the overall degree of pulmonary fibrosis in the patient at that time point.

[0029] (2) Calculation of the average annual growth rate of fibrosis A trend fitting method based on multi-time point data was used to model the time series data of fibrosis ratio changes over time, and then the average rate of change of fibrosis ratio per unit time was estimated and uniformly converted into an annual growth rate as the patient's average annual fibrosis growth rate.

[0030] Alternatively, the average annual fibrosis growth rate during the overall follow-up period can be obtained by calculating the change in the proportion of fibrosis between the baseline time point and each follow-up time point, and then weighting the change in the proportion of fibrosis with the corresponding time intervals.

[0031] The formula for the average annual fibrosis growth rate (unit: % / year) is as follows:

[0032]

[0033]

[0034] In the formula, For the first The difference in the percentage of fibrotic volume between the follow-up and baseline. For time intervals, Baseline CT scan time (in months). For the first Time interval for follow-up CT scans (unit: month).

[0035] The above methods can effectively reduce the impact of single measurement errors and uneven follow-up time on the growth rate calculation results, making the obtained growth rate index more stable and clinically reliable.

[0036] This invention constructs a method for modeling and predicting the progression of pulmonary fibrosis based on the fibrosis volume percentage index extracted from time-series CT images. This method is used to characterize the evolution of the degree of fibrosis in patients over time and to predict the fibrosis status at future time points.

[0037] (1) Modeling of fibrosis progression trajectory Based on time-series data of the change in the fiber ratio over time, a time-fiber ratio mapping relationship progression model is constructed. This model is used to characterize the overall trend of the change in the fiber ratio over time, and its functional form is not limited.

[0038] In different implementations, the time-fiber ratio mapping progression model may take one of the following forms, but is not limited to: For each patient, set For the first Time between CT scans (unit: months) This represents the percentage of fibrous volume at the corresponding time point.

[0039] This yields a time series data set:

[0040] A parameterized trend model is used to describe the continuous change in the proportion of fibrosis. For example, a small number of parameters can be used to characterize the patient's baseline fibrosis burden and its rate of progression over time, thereby achieving an effective characterization of disease progression while ensuring computational efficiency.

[0041] The model parameters are obtained by fitting or estimating follow-up data at multiple time points, enabling the model to reflect the characteristics of fibrosis progression at the individual patient level.

[0042] When the fibrosis ratio shows an approximately linear growth trend, a linear trend model is used for fitting; when the fibrosis ratio shows a significantly accelerating growth trend over time, an exponential growth model or a nonlinear growth model is used for fitting. The fitting errors of different models (such as mean squared error (MSE) or coefficient of determination) are compared. The model with the best fit was selected as the predictive model for the progression of fibrosis in this patient.

[0043] The functional relationship between the fiberization ratio and time was established using the least squares fitting method:

[0044] In the formula, Indicates time Predicted fiberization rate at any given time; The baseline fibrosis percentage parameter for patients; This is a parameter representing the rate of change of the proportion of fibrosis per unit time, and can be used to calculate the average annual fibrosis growth rate of a patient. One method.

[0045] Piecewise modeling of the progression function is used to characterize the changes in the rate of fibrosis progression at different time stages. For example, a piecewise linear model can be used to characterize the changes in the rate of fibrosis progression at different stages of the disease:

[0046] In the formula, This serves as a dividing point in the progression of the disease.

[0047] Nonlinear growth models are used to describe the accelerating or slowing characteristics of fibrosis at different disease stages. For example, the Logistic growth model can be used to describe the gradual stabilization of fibrosis progression at different disease stages.

[0048] For example, the exponential growth model characterizes the accelerated progression of disease:

[0049] in, The baseline fibrosis percentage; The disease progression growth coefficient is obtained by nonlinear regression estimation of data from multiple time points.

[0050] Time series models based on probability or statistical inference are used to characterize the uncertainty of fibrosis progression. For example, Markov time series models based on state transition probabilities can be used to describe the probability of transitions between different fibrosis progression states in patients.

[0051] The progression model employs a simplified trend function form, obtained by fitting or estimating multi-time-point follow-up data using a small number of parameters to characterize the patient's baseline fibrosis burden and its rate of progression over time. This allows the model to reflect the individual-level characteristics of fibrosis progression. The model parameters are obtained by fitting or estimating multi-time-point follow-up data, enabling the model to reflect the individual-level characteristics of fibrosis progression.

[0052] (2) Grouping modeling method based on progress rate Based on the aforementioned calculation of the average annual fibrosis growth rate, patients were divided into different progression types, including a rapid progression group, a slow progression group, and a stable group: the rapid progression group ( 10% / year), slow progression group (5) 10% / year), stable group ( 5% / year), such as Figure 2 As shown; using the KM survival analysis method, survival curves were established for the three groups of patients to verify the association between the rate of progression and overall survival (p<0.05). Based on this, corresponding progression models were constructed for patients with different progression types, allowing patients within the same group to share similar progression pattern parameters, thereby reducing the impact of individual measurement noise on model stability and improving the model's generalization ability on new patients, such as... Figure 3 As shown, this group-driven modeling approach can more accurately reflect the evolution of fibrosis along different disease progression pathways.

[0053] In practical applications, the system can receive the patient's baseline CT images and target prediction time points as input, and output the predicted fibrosis volume percentage at the target time point based on the progression model. Simultaneously, the system can also output the patient's corresponding progression group information and the survival risk or prognostic reference information associated with that group, providing quantitative support for clinical decision-making.

[0054] like ≥10 % / year, patients enter the rapid progression group, using a highly weighted accelerated predictive parameter, i.e., the exponential growth model: Predicting the progression of fibrosis. Alternatively, a linear model fitted using the least squares method or the annual average fibrosis growth rate can be used. or Predicting the progression of fibrosis.

[0055] If 5 10% / year, patients enter the slow progression group, using a linear model fitted with least squares or the average annual fibrosis growth rate. or Predicting the progression of fibrosis; like 5% / year, patients enter the stable group, using a linear model fitted by least squares or the average annual fibrosis growth rate. or Predicting the progression of fibrosis.

[0056] In addition, if the patient has progressed to an advanced or plateau phase, [the following can be used] Predicting the progression of fibrosis.

[0057] Predictive Output: Input the patient's baseline CT images and the target prediction time (e.g., 6 months, 12 months…36 months). The trained model outputs the predicted fibrosis volume percentage, progression group, and corresponding survival prognosis at the target time point as a reference. Figure 4 As shown.

[0058] The accuracy of image prediction was assessed by the overlap of fibrotic areas between predicted CT images and actual follow-up CT images. The reliability of the association between progression grouping and survival prognosis was verified using the fibrosis progression status and survival outcome during clinical follow-up as the gold standard. The consistency between the fibrosis percentage predicted by linear regression at each time point and the actual measured values ​​was verified.

[0059] Retrospective analysis revealed that the hazard ratio for survival prediction based on fibrosis progression grouping (rapid progression / slow progression / stable progression) was 1.606 (1.199 - 2.150, p = 0.0015), making it the only statistically significant survival predictor. Fibrosis progression grouping was an independent risk factor for poor patient survival. However, the proportion of whole-lung fibrosis at a single node and the proportion of fibrosis in a specific lung lobe did not meet the p-value requirement of p < 0.05, indicating no statistical significance, and therefore could not be used as survival predictors.

[0060]

[0061] The beneficial effects are as follows: 1. Achieve temporal and continuous prediction: Construct a time-fibrosis ratio mapping model, build a time sequence based on multiple CT scans, break through the limitations of traditional static assessment, and accurately predict fibrosis progression for up to 36 months, providing a basis for long-term treatment planning.

[0062] 2. More precise quantitative grouping: Through AI-assisted segmentation and growth rate calculation, IPF patients were grouped into three levels of progression. The average annual fibrosis rates of the three groups were 13.23%, 5.83%, and 1.49%, respectively. The grouping results were significantly correlated with survival prognosis (p<0.05), providing a quantitative basis for stratified treatment.

[0063] 3. Improve the objectivity and efficiency of assessment: Replace manual visual assessment, reduce subjective bias, automate segmentation, calculation, grouping and prediction, and reduce clinical workload.

[0064] 4. Expanding clinical application scenarios: It can monitor early signs of acute exacerbations, assess the efficacy of antifibrotic drugs, guide the adjustment of treatment plans, and establish a correlation between "imaging-growth rate-prognosis" to provide both doctors and patients with a more comprehensive understanding of the disease.

[0065] 5. High applicability and transferability: It relies solely on CT image data, requires no additional complex clinical indicators, and is easy to promote in clinical practice; the method can be transferred to other related diseases that cause pulmonary interstitial fibrosis, and has a wide range of application scenarios.

[0066] 6. Significantly improved disease course adaptability: The constructed model matches the decision logic, which can accurately adapt to the evolution characteristics of IPF throughout the entire disease course, such as stable progression, accelerated progression, and late plateau stage. This effectively avoids underestimating the disease risk of patients with rapid progression, and significantly improves the accuracy of prediction and clinical applicability.

Claims

1. A method for predicting the progression of pulmonary fibrosis based on time-series CT images, characterized in that, Includes the following steps: Baseline, follow-up CT images, and survival follow-up information of multicenter IPF patients were collected to construct a time-series sequence of CT images; The time-series CT images were preprocessed, followed by segmentation of the whole lung region and the fibrotic region. All voxels in the whole lung region and all voxels in the fibrotic region were counted. The whole lung volume and fibrotic volume were calculated by combining the corresponding voxel spatial resolution and normalized to obtain the percentage of fibrotic volume to the whole lung volume at different time points, forming time-series data of fibrosis ratio changing over time. The annual average fibrosis growth rate was calculated based on the time-series data of fibrosis ratio changing over time or the change in fibrosis ratio between the baseline time point and each follow-up time point. Based on time-series data of fibrosis proportion changing over time, a time-fibrosis proportion mapping model was constructed. A progression type map was plotted based on the annual average fibrosis growth rate. The KM survival analysis method was used to establish survival curves for multiple progression types. Corresponding progression models were constructed for patients with different progression types, so that patients in the same group could share similar progression pattern parameters. Based on the baseline, follow-up CT images, survival follow-up information, percentage of fibrosis volume to total lung volume, time-series data of fibrosis proportion changing over time, annual average fibrosis growth rate, progression type map, survival curves for multiple progression types, and progression models corresponding to different progression types, a pulmonary fibrosis progression prediction based on time-series CT images was constructed for multicenter IPF patients. The patient's baseline, follow-up CT images, and survival follow-up information are input into the time-series CT image-based pulmonary fibrosis progression prediction system. The system outputs the percentage of predicted fibrosis volume in the total lung volume, the progression group, and the corresponding survival prognosis.

2. The method for predicting the progression of pulmonary fibrosis based on time-series CT images as described in claim 1, characterized in that, The survival follow-up information includes survival status and follow-up cutoff time.

3. The method for predicting the progression of pulmonary fibrosis based on time-series CT images as described in claim 1, characterized in that, The preprocessing includes nonlinear denoising and image alignment.

4. The method for predicting the progression of pulmonary fibrosis based on time-series CT images as described in claim 1, characterized in that, The time-fiber ratio mapping model includes: A parametric trend model is used to describe the continuous change process of the fiber ratio. A piecewise modeling progression function is used to characterize the changes in the rate of fibrosis progression at different time stages; Nonlinear growth models are used to describe the acceleration or deceleration characteristics of fibrosis at different stages of the disease. Time series models based on probability or statistical inference are used to characterize the uncertainty of fibrosis progression.

5. The method for predicting the progression of pulmonary fibrosis based on time-series CT images as described in claim 4, characterized in that, The parameterized trend model adopts a simplified trend function form, which uses a small number of parameters to characterize the patient's baseline fibrosis burden and its rate of progression over time. It is obtained by fitting or estimating follow-up data at multiple time points, so that it can reflect the fibrosis progression characteristics at the individual patient level.

Images