System and method for identifying feature in an image of a subject

Inventors

Helm, Patrick A. • Siewerdsen, Jeffrey H. • Uneri, Ali • Jones, Craig K. • HUANG, Yixuan • Zhang, Xiaoxuan

Abstract

A method and system is disclosed for analyzing image data of a subject. The image data can be collected with an imaging system in a selected manner and/or motion. The image data may include selected overlap and be acquired with an imaging system that generates a plurality of perspectives for more than one location. An automatic system and method may then define or identify various features and/or allow for registration for alternative image data.

Core Innovation

The invention relates to detecting and classifying a feature in an image using an imaging system that generates a plurality of individual image projections. The plurality of individual image projections comprise a first set from a first view of a subject and a second set from a second view of the subject, wherein the first view is different than the second view. Proposed regions for the feature are determined in a first feature map associated with the first view and a second feature map associated with the second view.

The proposed regions are combined across the first feature map and the second feature map, and the feature is classified while determining a position of the feature in each of the first view and the second view based on combining the proposed regions. The output includes a first label identifying the feature and the position of the feature in the first output view, and a second label identifying the feature and the position of the feature in the second output view. The concept is implemented as a method and as a corresponding system comprising a processor module configured to execute instructions.

In described embodiments, the plurality of individual image projections are acquired using an x-ray imaging system with a movable source and detector and a gantry/cart arrangement, and a slotted filter that generates multiple fan-beam projections that exhibit a parallax effect. The projections are stitched into intermediate films and a reconstructed long film. In described embodiments, automated feature detection and labeling is performed using multi-slot and/or multi-view deep learning pipelines, including proposed region generation, combining operations, and an optional Bi-LSTM confirmation module to enforce sequential anatomical ordering for features such as vertebrae and implants.

The described content also includes registration of intraoperative multi-view images to preoperative 3D data, including subject registration using multi-scale masking with synthetic slot image generation and optimization, followed by device registration using a known-component device model and gradient correlation optimization. The described content states improved accuracy over rigid transformation for implanted device positioning in the presence of metallic interference.

Claims Coverage

The partial content identifies two independent claims: clm-00001 (method) and clm-00012 (system). Each independent claim includes a set of inventive features centered on multi-view projection-based feature detection, proposed-region combination across view-associated feature maps, and labeled output of feature positions in both output views.

Across both independent claims, coverage centers on acquiring plurality of individual image projections from different views, mapping proposed regions into view-associated feature maps, combining those proposed regions to jointly enable classification and per-view position determination, and outputting labeled results for both output views that identify the feature and its position.

Stated Advantages

Documented Applications