Deep CNN Tumor Segmentation in PET-CT Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated tumor segmentation methods face challenges in accurately delineating tumors from PET scans due to ambiguous boundaries, low resolution, and low contrast in PET-CT images, leading to bias and variance in computing standardized uptake values (SUVs), which affects tumor detection and treatment assessment.
Innovation Solution
The use of a convolutional neural network architecture that incorporates two-dimensional and three-dimensional segmentation models, with residual blocks and skip connections, to generate standardized images and segmentation masks from PET and CT/MRI scans, enabling accurate tumor segmentation and metabolic burden assessment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional automated tumor segmentation methods are used on PET-CT images, then the process is simple and fast, but the segmentation accuracy is low due to ambiguous boundaries, low resolution, and low contrast
Solution Approach 1:
The patent applies segmentation by dividing the tumor segmentation task into multiple processing stages using a deep convolutional neural network with residual blocks. The network processes images through multiple layers that progressively extract features and refine segmentation boundaries, effectively breaking down the complex task of delineating ambiguous tumor boundaries from PET-CT images into manageable computational steps that improve accuracy
Solution Approach 2:
The patent employs three-dimensional convolutional neural networks that process volumetric PET-CT data in 3D space rather than traditional 2D image processing. This dimensional transition enables the model to capture spatial relationships and metabolic information across multiple slices, improving tumor segmentation accuracy by considering the three-dimensional structure of tumors and their boundaries
2Measurement precision
If deep convolutional neural networks with residual blocks are used, then segmentation accuracy improves, but computational complexity and processing time increase
Solution Approach 1:
The patent implements preliminary action through pre-processing steps that standardize input PET-CT images before they enter the deep neural network. The system performs normalization, registration, and feature enhancement on the input images, which prepares the data in advance to reduce the computational burden during the actual segmentation process, thereby balancing accuracy with processing efficiency
Solution Approach 2:
The patent uses residual blocks that incorporate skip connections, which can be viewed as copying feature maps from earlier layers directly to later layers. This copying mechanism allows the network to preserve important low-level features while adding complex high-level processing, improving segmentation accuracy without requiring excessively deep networks that would increase processing time
3Measurement precision
If standard PET scan resolution is used, then image acquisition is fast and simple, but tumor boundary delineation is ambiguous and inaccurate
Solution Approach 1:
The patent introduces an intermediary deep convolutional neural network that acts as a mediator between the low-resolution PET scan data and the required high-precision tumor boundary delineation. The network processes the standard-resolution images through multiple convolutional layers with residual connections, effectively enhancing boundary definition and segmentation accuracy without requiring changes to the PET scanner hardware or acquisition protocols
Data Source
AI summary
The present disclosure relates to techniques for segmenting tumors with positron emission tomography (PET) using deep convolutional neural networks for image and lesion metabolism analysis. Particularly, aspects of the present disclosure are directed to obtaining a PET scans and computerized tomography (CT) or magnetic resonance imaging (MRI) scans for a subject, preprocessing the PET scans and the CT or MRI scans to generate standardized images, generating two-dimensional segmentation masks, using two-dimensional segmentation models implemented as part of a convolutional neural network architecture that takes as input the standardized images, generating three-dimensional segmentation masks, using three-dimensional segmentation models implemented as part of the convolutional neural network architecture that takes as input patches of image data associated with segments from the two-dimensional segmentation mask, and generating a final imaged mask by combining information from the two-dimensional segmentation masks and the three-dimensional segmentation masks.