Hierarchical CT Parsing for Organ Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for segmenting multiple structures in whole body CT scans are organ-specific, struggle with topological changes due to disease or organ movement, and are computationally expensive, making them inaccurate and time-consuming.
Innovation Solution
A hierarchical parsing method that uses a Discriminative Anatomical Network (DAN) and database-guided segmentation module to detect anatomic landmarks and segment multiple organs efficiently, incorporating contextual information and handling topological changes, allowing for unified segmentation and detection of organs and landmarks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If non-rigid registration techniques are used for segmenting multiple organs, then the method can handle organ deformation, but the computational cost increases significantly leading to long processing times
Solution Approach 1:
The patent segments the segmentation process into two distinct phases: (1) a fast rigid registration phase that aligns the atlas to the patient anatomy using only rigid transformations, and (2) a subsequent organ-specific segmentation phase that handles deformation locally for each organ. This segmentation avoids the computational burden of full non-rigid registration while still achieving accurate organ segmentation by treating each organ's deformation independently after the rigid alignment.
Solution Approach 2:
The patent extracts and removes the computationally expensive non-rigid registration component from the segmentation pipeline. Instead of applying non-rigid registration to the entire atlas, the method extracts only the necessary rigid transformation and performs organ-specific segmentation separately, thereby eliminating unnecessary computational steps while preserving the ability to handle organ deformation where needed.
2Measurement precision
If non-rigid registration is applied to handle topological changes in organ boundaries, then accuracy improves for abnormal organs, but the computational expense increases
Solution Approach 1:
The patent applies local quality by performing segmentation and deformation handling locally for each organ rather than applying global non-rigid registration to the entire anatomy. Each organ is segmented independently after rigid alignment, allowing the method to handle topological changes and abnormalities in specific organs without incurring the computational cost of global non-rigid registration. This localized approach maintains precision for abnormal organs while reducing overall computational energy consumption.
3Measurement precision
If organ-specific segmentation methods are used, then accuracy for specific organs improves, but the method becomes difficult to transfer to other organs
Solution Approach 1:
The patent implements universality by using a single rigid registration framework that can be applied to segment multiple different organs. The method registers the atlas to patient anatomy using rigid transformation, then performs segmentation for various organs (heart, liver, kidneys, spleen, bladder, prostate) using the same foundational approach. This unified rigid registration framework makes the method transferable to different organs and anatomical regions while maintaining accuracy through organ-specific segmentation parameters and constraints.
Data Source
AI summary
A method and apparatus for hierarchical parsing and semantic navigation of a full or partial body computed tomography CT scan is disclosed. In particular, organs are segmented and anatomic landmarks are detected in a full or partial body CT volume. One or more predetermined slices of the CT volume are detected. A plurality of anatomic landmarks and organ centers are then detected in the CT volume using a discriminative anatomical network, each detected in a portion of the CT volume constrained by at least one of the detected slices. A plurality of organs, such as heart, liver, kidneys, spleen, bladder, and prostate, are detected in a sense of a bounding box and segmented in the CT volume, detection of each organ bounding box constrained by the detected organ centers and anatomic landmarks. Organ segmentation is via a database-guided segmentation method.