Method and system for determining irradiation dose

A method for determining an irradiation dose, comprising:

• determining a pixel having a biological feature in a region of interest in a radiotherapy simulated locating image by using a retrospective label, wherein the radiotherapy simulated locating image includes computed tomography imaging, nuclear magnetic resonance imaging, and positron emission tomography imaging, and the biological feature comprises the glucose metabolism level, oxygen content level, angiogenesis conditions in the tumor, and the magnitude of complications risk in normal tissues;
• wherein the determining a pixel having a biological feature in a region of interest in a radiotherapy simulated locating image by using a retrospective label comprises:
• carrying out an image registration between images reflecting biological features of the region of interest and the radiotherapy simulated locating image by retrospective analysis, and then labeling corresponding pixels in the region of interest of the radiotherapy simulated locating image base on whether pixels in the images reflecting biological features of the region of interest have a biological feature of interest, wherein labeling positive pixel with 1 while labeling negative pixel with 0;
• all pixels in a ROI of a certain region of interest are defined as a set P={plabel 1, plabel 2, . . . , plabel n}, where n represents a number of pixels, and label represents a mark of a pixel, where 1 refers to positive, and 0 refers to negative;
• extracting the local radiomics feature based on the pixel having the biological feature, wherein the local radiomics feature includes a grayscale histogram intensity, a tumor shape feature, a textural feature, a Laplacian of Gaussian (LoG) filtering feature, and a wavelet feature;
• wherein the extracting the local radiomics feature based on the pixel having the biological feature comprises:
• traversing all of the pixels in each region of interest of the radiotherapy simulated locating image, and calculating the local radiomics feature values of δ×δ×δ neighboring windows of three-dimensional images or β×β neighboring windows of two-dimensional images which surrounding each pixel by using each pixel as the center, in which δ is an odd number greater than or equal to 3, and the δ value is taken as 3, 5, 7, and 9 respectively each time for the calculation of local radiomics features, such a selection in modeling and verification steps enabling recognition of a δ value with the optimum accuracy;
• then a 1×N dimension feature vector of Flabel i={f1(plabel i), f2(plabel i), . . . , fj(plabel i)} is obtained for each pixel at a specific δ value, where j∈{1, . . . , N},i∈{1, . . . , n}, label∈{1,0} ;
• for the pixel plabel i, N feature values are extracted, which are f1(p), f2(pi), . . . , fj(pi) respectively, and are marked as a vector Flabel I, therefore, for all pixels in the ROI, the feature set is {Flabel i, . . . , Flabel n}, label∈{1,0};
• each pixel serves as one sample with N-dimensional local radiomics features;
• for pixels on the boundary, symmetric filling is adopted, and the value of the fill pixel is a mirror reflection of the boundary pixel;
• acquiring the local radiomics feature to be measured;
• identifying a positive region having the local radiomics feature to be measured based on the local radiomics feature;
• wherein the identifying a positive region having the local radiomics feature to be measured based on the local radiomics feature further comprises:
• screening the local radiomics features to determine an optimal feature subset G={f1(p), f2(p), . . . , fk(p)};
• each labeled pixel and its local radiomics features serve as one sample to participate in feature selection, and both positive and negative pixels participate in this process;
• establishing a supervised machine learning model based on the optimal feature subset;
• verifying an identification accuracy of the supervised machine learning model based on data of a verification set, and selecting a feature extraction window δopt that makes the identification accuracy optimal;
• and
• identifying a positive region having the local radiomics features to be measured based on the supervised machine learning model;
• for a new locating image, for which a local radiomics feature-guided radiotherapy plan is to be formulated, the local radiomics features described in the set G are extracted pixel by pixel through a neighboring window having the size of δopt;
• and the feature parameters of each pixel are input into the supervised machine learning model to output the biological-function expression label of the current pixel;
• the pixels being positive for the same biological function are labeled with a unified color, and a biological-function positive region being displayed on each tomography image;
• performing three-dimensional reconstruction for the peripheral boundary of the positive region to determine a three-dimensional image, wherein the three-dimensional image is a three-dimensional image displaying the biological features;
• and
• determining irradiation doses at different locations in different regions based on the three-dimensional image.