Joint CT-MR Reconstruction From Sparse Single-Modality Scans
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging techniques involving CT and MR scans are lengthy, expose patients to high radiation, and face challenges in accurately registering images due to anatomical changes over time, complicating treatment planning and delivery.
Innovation Solution
The Adaptive Neural Representation (ANR) method uses a multi-layer perceptron neural network with an anatomy-adaptive layer to simultaneously reconstruct CT and MR images from sparse measurements from a single modality, leveraging a patient-specific model to adjust to new anatomical data quickly and reduce reliance on large datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If sequential acquisition of CT and MR images is performed, then both imaging modalities can be obtained, but acquisition time becomes prohibitively long
Solution Approach 1:
The patent combines CT and MR image acquisition into a single simultaneous measurement process. The system acquires both modalities from the same patient position and anatomical state in one procedure, eliminating the time loss associated with sequential scanning while maintaining the benefits of both imaging modalities.
Solution Approach 2:
The imaging system is designed to perform multiple imaging functions simultaneously - acquiring both CT and MR data from a single measurement session. This multi-functional capability allows the system to provide comprehensive imaging information without requiring separate dedicated scanning sessions.
2Measurement precision
If dense sampling is used in measurement space, then image quality is improved, but radiation dose and reconstruction time increase
Solution Approach 1:
The patent employs partial sampling techniques where only a subset of measurement data is required to achieve high-quality image reconstruction. By using partial rather than complete sampling, the system maintains image quality while significantly reducing the radiation dose and computational resources needed compared to traditional dense sampling approaches.
Solution Approach 2:
The system uses deep learning models to generate synthetic measurement data that replicates the information content of dense sampling without actually acquiring all the data. This copying approach allows high-quality image reconstruction from sparse measurements, reducing radiation exposure while maintaining diagnostic quality.
3Reliability
If CT and MR images are acquired independently, then each modality can be optimized, but anatomical changes during time gap complicate treatment
Solution Approach 1:
The patent performs preliminary simultaneous acquisition of both CT and MR images from the same patient position before any treatment intervention. This preliminary action ensures that both modalities capture the anatomical structure in its original state, eliminating subsequent registration problems caused by anatomical changes that would occur if images were acquired sequentially over time.
4Adaptability or versatility
If image registration is performed to combine CT and MR images, then multi-modal information can be integrated, but additional uncertainties are introduced
Solution Approach 1:
The system performs preliminary simultaneous acquisition of CT and MR images from the identical patient position and anatomical state. Because both images are obtained at the same time from the same position, no post-acquisition registration is needed - the images are already spatially aligned, eliminating registration uncertainties while enabling multi-modal integration.
Solution Approach 2:
The patent uses a common reference coordinate system and anatomical landmark framework as an intermediary that naturally aligns both CT and MR images during simultaneous acquisition. This intermediary reference framework ensures consistent spatial positioning without requiring complex registration algorithms, thereby reducing uncertainty while enabling reliable multi-modal integration.
Data Source
AI summary
A method for medical imaging includes performing a single-modality scan of a subject using a single-modality imaging device to acquire sparse measurements, wherein the single-modality is either computed tomography (CT) or magnetic resonance imaging (MRI); and simultaneously reconstructing both CT and MR image pairs from the sparse single-modality measurements using a multi-layer perceptron (MLP) neural network; wherein initial weights of the MLP are learned from a pair of pre-treatment CT and MR images of the subject.