Hierarchical Neural Network for Anatomical Structure Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The interpretation of medical images, particularly ultrasound images, is challenging due to artefacts, noise, and high intra-user and inter-user variability, even among specialists, necessitating computer-assisted detection and identification of anatomical structures.
Innovation Solution
A computer-implemented method using a hierarchical anatomical identification scheme that applies two neural networks sequentially to detect larger-sized and smaller-sized anatomical structures in medical images, allowing for robust and precise detection of anatomical features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single neural network is used to detect all anatomical structures, then the device complexity is reduced, but the detection precision decreases for smaller anatomical structures
Solution Approach 1:
The patent divides the detection task into multiple neural networks, each specialized for detecting anatomical structures of specific size ranges. This segmentation allows each network to optimize its detection capabilities for particular structure sizes, thereby improving overall detection precision without requiring a single overly complex network to handle all sizes equally well.
Solution Approach 2:
The patent introduces a hierarchical dimension to the detection system by organizing neural networks into multiple levels based on anatomical structure size. This hierarchical arrangement adds a dimensional aspect (size-based hierarchy) to the detection process, allowing the system to systematically address structures from largest to smallest with appropriately scaled detection models.
2Measurement precision
If multiple neural networks are applied sequentially, then the detection precision improves, but the processing time increases
Solution Approach 1:
The patent implements preliminary detection at each hierarchical level before proceeding to the next. By detecting and localizing larger anatomical structures first, the system can then focus subsequent detection efforts on smaller structures within the identified regions, eliminating the need to search the entire image at all scales and thereby reducing overall processing time.
Solution Approach 2:
The sequential application of multiple neural networks is segmented into hierarchical stages, where each stage processes only the relevant size range of anatomical structures. This segmentation of the processing pipeline allows the system to maintain high detection precision through multiple specialized networks while reducing total processing time by avoiding redundant computations across all networks for every image region.
3Measurement precision
If the search area is reduced by cropping, then the detection precision for smaller structures improves, but the device complexity increases
Solution Approach 1:
The patent uses preliminary detection of larger anatomical structures to define cropping regions for subsequent detection of smaller structures. This preliminary action of identifying broader anatomical contexts first allows the system to strategically crop and focus computational resources on relevant regions, improving detection precision for smaller structures while managing complexity through intelligent region selection rather than processing the entire image at all scales.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a computer-implemented method for automatically detecting anatomical structures (3) in a medical image (1) of a subject, the method comprising applying an object detector function (4) to the medical image, wherein the object detector function performs the steps of: (A) applying a first neural network (40) to the medical image, wherein the first neural network is trained to detect a first plurality of classes of larger-sized anatomical structures (3a), thereby generating as output the coordinates of at least one first bounding box (51) and the confidence score of it containing a larger-sized anatomical structure; (B) cropping (42) the medical image to the first bounding box, thereby generating a cropped image (11) containing the image content within the first bounding box (51); and (C) applying a second neural network (44) to the cropped medical image, wherein the second neural network is trained to detect at least one second class of smaller-sized anatomical structures (3b), thereby generating as output the coordinates of at least one second bounding box (54) and the confidence score of it containing a smaller-sized anatomical structure.