Screening method for quantitative prognostic features of digital pathology images

Abstract

The invention discloses a method for screening quantitative features of prognosis of digital pathological images, which includes obtaining quantitative features a (0) ; using Cox proportional hazards algorithm, machine learning model and Pearson correlation coefficient to quantify feature a (0) Filter to get the feature vector a (3) ;Take the eigenvector a (3) Construct a Cox proportional hazards model and update the eigenvector a using the Cox proportional hazards model (3) , execute this step again, and then update the obtained feature vector a (3) Input the Cox proportional hazard model, output the characteristic risk coefficient, and calculate the proportional hazard value. Discretize the risk coefficient corresponding to each feature in the feature vector a to obtain the risk assessment score of each feature; divide the features whose assessment score is greater than the cutoff value into group Q, and divide the rest into group N; use group The corresponding features in Q are used as prognostic quantification features.

Description

technical field [0001] The invention relates to a medical image extraction technology, in particular to a screening method for the prognosis quantification feature of a digital pathological image. Background technique [0002] The occurrence and development of cancer is the result of the interaction between cancer cells and the tumor microenvironment. The changes in the type, quantity or shape of cells in the tumor stroma have important medical guiding significance. For example, lymphocytic infiltration in breast cancer generally has a better prognosis, whereas the presence of tumor-associated fibroblasts is associated with a poor prognosis. In routine pathological work, changes in cellular components and extracellular matrix in the tumor stroma are generally described qualitatively. Based on digital pathological image analysis, different components in the interstitium can be automatically segmented and quantitatively or qualitatively studied. [0003] In quantitative rese...

AI Technical Summary

Problems solved by technology

[0004] At present, when medical researchers analyze digital pathological images of patient prognosis, they only need to analyze the morphological quantitative features closely related to tumor recurrence to find the cause of tumor recurrence; , often includes all morphological quantitative features, and the amount of data is relatively large. If the morphological quantitative features closely related to tumor recurrence are manually found and screened from the public database, the workload of researchers is very large, and During manual recognition, due to the visual fatigue caused by browsing a large amount of information, the accuracy of recognition is difficult to guarantee

Images