Method for detecting jugular vein ball socket bone wall deficiency in temporal bone CT image

Abstract

The invention provides a method for detecting jugular vein ball socket bone wall deletion abnormity in a temporal bone image. The method can be used for computer-aided pulsatile tinnitus diagnosis. According to the method, a small amount of rare abnormal data is fully utilized, abnormal prior information is introduced into a network model, abnormal features are highlighted, a difference between the abnormal features is increased, and the overall detection accuracy is improved. Through a targeted data enhancement mode and an innovative data use means, a reasonable proxy task is designed, so that the network model gives full play to a advantage of being good at prediction, and detection accuracy is improved. Compared with an existing method, the method provided by the invention does not depend on a large amount of labeled data any more, and has advantages of being easier to implement and popularize.

Description

technical field [0001] The invention relates to the field of medical image processing, in particular to a method for detecting the abnormality of bone wall loss of the jugular vein ball socket image in temporal bone CT. Background technique [0002] Tinnitus is manifested as the abnormal perception of sound by the human body without external stimuli. It is a common and frequently-occurring disease in otology. Among them, pulsatile tinnitus (Pulsatile Tinnitus, PT) accounts for about 4%-10% of tinnitus patients. In the past ten years, with economic development and people's emphasis on health, the demand for diagnosis and treatment of patients has surged. With the popularization of temporal bone imaging technology, the imaging method for pulsatile tinnitus has become an important technical means. [0003] Temporal bone CT is the main method for diagnosing osseous lesions in pulsatile tinnitus. Temporal bone CT can display multiple fine structures such as the middle ear and i...

AI Technical Summary

Problems solved by technology

For the abnormal detection of jugular vein ball-socket bone wall loss, there may be the following problems: (1) The jugular vein ball-socket target size is small in temporal bone CT, and the bone wall loss lesion target is even smaller, which makes target detection difficult and a large number of false positive samples are likely to be generated; ( 2) Abnormal samples are far less than normal samples, the categories are extremely unbalanced, features are difficult to extract, and categories are difficult to distinguish

Anomaly detection based on self-encoder has strong network generalization ability. When an abnormal image is input, the decoded image is very close to the input image, and it is difficult to distinguish between normal and abnormal.

Codec networks are usually used in unsupervised tasks without the guidance of abnormal data

Images