Method and device for evaluating image annotation diagnosis quality

Abstract

The invention provides a method and a device for evaluating image annotation diagnosis quality. The method comprises the following steps: randomly displaying an image of a case corresponding to a subject selected by a doctor; extracting a focus contour delineated by a doctor, calculating the similarity between the focus contour and the focus contour of the same image in the gold standard, and if the similarity is greater than a set threshold value, judging that the focus contour delineation is correct; extracting labels and global labels which are labeled by doctors and are related to focus types, properties and follow-up visit suggestions, and respectively matching the labels with corresponding labels in gold standards to judge whether each label is correct or not; scoring is performed according to the lesion contour and the judgment result of each label, the score of each question is obtained, and the evaluation score of the doctor is obtained after summarizing. According to the invention, quantitative evaluation can be carried out on the ability of a doctor to sketch a lesion contour and label, and the problem that the prior art only emphasizes theoretical examination and is difficult to carry out all-around quantitative evaluation on the labeled diagnosis quality of the doctor is solved.

Description

technical field [0001] The invention relates to the technical field of medical imaging, in particular to a method and device for evaluating the diagnostic quality of image annotation. Background technique [0002] It has been more than 100 years since the discovery of X-rays. During these 100 years, medical imaging technology and related industries have developed rapidly, and medical imaging has become an indispensable tool in the process of clinical diagnosis and treatment. With the aid of medical imaging, early screening can detect nodules, tumors and other diseases as early as possible; in clinical diagnosis, it can assist doctors to effectively locate the location of lesions, determine the type of lesions, and help patients' treatment. Accurate and rapid film reading and diagnosis are one of the essential skills of radiologists. Doctors with rich experience and high qualifications can often make more accurate diagnoses based on images, so that patients have a greater cha...

AI Technical Summary

Problems solved by technology

[0003] At present, the exams for imaging doctors mostly adopt the standardized training method, and the questions focus on theory, so it is difficult to comprehensively evaluate the doctor's diagnostic level and labeling quality

Although senior doctors can estimate the labeling and diagnosis level of junior doctors by looking at the results of some images labeled by junior doctors, this method is often time-consuming and cannot be quantitatively evaluated

In addition, in order to evaluate the auxiliary diagnosis effect of AI software, it is often necessary to manually count the changes in the doctor's diagnosis results before and after the application of AI assistance, which takes a lot of time

Images