Deep Learning Image Reconstruction for High-Resolution DOT
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current diffuse optical tomography (DOT) systems face challenges in spatial resolution and reconstruction complexity, limiting their effectiveness in characterizing smaller breast lesions and requiring labor-intensive preprocessing, which hinders clinical adoption.
Innovation Solution
A deep learning-based approach using a machine learning model with a 3D convolutional neural network (U-Net) architecture enhances image resolution and reduces noise by leveraging multimodal data, including structural and functional imaging, and incorporates a prior-weighted loss function to improve lesion characterization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative optimization with FEM numerical model is used for DOT reconstruction, then reconstruction accuracy is improved, but computational time and complexity increase significantly
Solution Approach 1:
The patent creates a digital twin of the physical imaging system by training a neural network on simulated data that replicates the forward model of light transport. This virtual copy allows rapid inference without repeated FEM simulations, achieving both accuracy and speed.
Solution Approach 2:
The system performs preliminary training offline using simulated data to learn the relationship between sensor measurements and tissue optical properties. During clinical use, this pre-learned knowledge enables rapid reconstruction through simple neural network inference, avoiding time-consuming iterative optimization.
2Stability of the object's composition
If regularization is applied to ensure convergence in solving the ill-posed DOT inverse problem, then solution stability is improved, but spatial resolution and contrast recovery deteriorate
Solution Approach 1:
The patent changes the approach from traditional Tikhonov regularization to data-driven learned regularization. The neural network automatically adapts regularization strength and spatial frequency characteristics based on the input data and task requirements, enabling high-resolution reconstruction without artificial constraint bias.
Solution Approach 2:
The patent replaces the mechanical/mathematical regularization framework with a neural network-based solution. Instead of applying explicit mathematical constraints that bias the solution, the system uses learned representations from data to achieve both stability and high resolution simultaneously.
3Measurement precision
If expert knowledge is used to optimize reconstruction performance for each individual case, then image quality is improved, but operational complexity and time consumption increase
Solution Approach 1:
The neural network system performs self-optimization by automatically adapting to different imaging scenarios through its learned parameters. It eliminates the need for expert manual tuning by internally optimizing reconstruction parameters based on the input data characteristics, making the system both high-performance and easy to operate.
Data Source
AI summary
An image may be reconstructed from sensor data, which may include optical imaging data such as diffusion optical tomography (“DOT”) data. The sensor data are received by a computer system. A machine learning model is accessed with the computer system, where the machine learning model includes a first subnetwork that receives sensor data as an input and generates an intermediate image as a first output, and a second subnetwork that receives the first output from the first subnetwork and generates an enhanced image as a second output. The sensor data are input to the machine learning model using the computer system, generating an enhanced image as an output. The enhanced image may have higher spatial resolution, reduced noise, or other improved image quality. Structural images may be passed as an additional input to the second subnetwork of the machine learning model to increase the spatial resolution of the enhanced image.