Cascaded Ensembled CNNs for Consistent PET/CT Lesion Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current lesion segmentation in 18F-FDG PET/CT images is labor-intensive, costly, and suffers from high inter-reader variability, making it infeasible in clinical routine.

Innovation Solution

A cascaded deep neural network approach using ensembled CNNs and a refiner model for lesion segmentation, which includes pre-processing to multiple intensity ranges and resolutions, and utilizes a combination of dice loss, cross-entropy loss, and sensitivity loss for improved accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual or computer-assisted lesion segmentation is used, then diagnostic accuracy can be maintained through expert interpretation, but the process becomes labor-intensive and costly

Engineering Contradiction:

Improvelesion segmentation accuracyVSAvoidsegmentation efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system enables automated self-service segmentation where the CNN model autonomously identifies and segments lesions without requiring manual intervention from nuclear medicine readers, thereby maintaining diagnostic accuracy while eliminating labor-intensive manual segmentation processes

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical manual segmentation process performed by human experts with an automated computer vision system based on convolutional neural networks, substituting human labor with an intelligent algorithm that can process images rapidly and consistently

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If manual lesion segmentation is performed by multiple readers, then diagnostic reliability can be cross-validated, but inter-reader variability increases and costs rise

Engineering Contradiction:

Improvediagnostic consistencyVSAvoidsegmentation process complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system ensures homogeneous segmentation results by applying the same trained CNN model consistently across all PET/CT images, eliminating the variability that arises from different human readers applying different interpretation criteria, thereby achieving uniform diagnostic standards

Inventive Principle:

Principle #33Homogeneity

3Productivity

If automated segmentation methods are implemented, then productivity and consistency are improved, but measurement precision may decrease due to loss of expert qualitative analysis

Engineering Contradiction:

Improvesegmentation throughputVSAvoidlesion characterization accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary automated segmentation to identify potential lesions and generate initial segmentation masks, which can then be used as a foundation for further analysis or minimal manual refinement, combining the speed of automation with the expertise of human readers when necessary

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250252568A1Systems and methods for image segmentation of pet/CT using cascaded and ensembled convolutional neural networks

Publication Date: 2025.08.07 SUBTLE MEDICAL INC

AI summary

A computer-implemented method is provided for segmentation of Positron emission tomography (PET)/computed tomography (CT). The method comprises: acquiring an original medical image including a PET image and CT image of a subject; transforming the original medical image into an input image with a predetermined resolution and a plurality of channels; processing the input image using an ensembled CNNs to output an intermediate segmentation mask; and taking the intermediate segmentation mask as input to a refiner model to output a final segmentation mask, where the final segmentation mask has a resolution same as the resolution of the original medical image.