TH-CD-204-10: Probability-Driven Sorting Method for Reconstructing Variation-Independent 4D-MRI

Citations Over Time

Abstract

Purpose: To develop a novel probability-driven sorting method that inherently incorporates breathing variations in the sorting process and 4D reconstruction. Methods: The probability-driven sorting method has two key steps: (a) determination of multiple principal breathing cycles that largely represent main variation patterns of the breathing signal. This will be accomplished by analyzing the breathing signal acquired during image acquisition using principal component analysis (PCA). The relative occurrence rate of each principal cycle will also be determined; (b) 4D image reconstruction based on the determined principal cycles. We will separately reconstruct a 4D-MRI for each principal cycle by optimally sorting k-space segments that best represent the principal cycle. Multi-cycle 4D reconstruction is impossible for most current 4D imaging techniques such as 4D-CT. It is however feasible for k-space reordered 4D-MRI due to the random assignment of phase/amplitude to k-space segments, as a Result of sampling scheme variations and breathing variations. To demonstrate the feasibility, we determined data completeness cures for the probability-driven sorting technique via simulation using breathing profiles from 30 cancer patients. It requires more repetitions to achieve full completeness relative to phase sorting, but is compatible to amplitude sorting. Results: We have tested the proposed method via simulation using 30 patients’ breathing profiles. We generated three PDFs per patient: one with all cycles (PDFA), one with a single cycle (PDFS), and one with the principal cycles (PDFP). We found that Dice's similarity coefficients comparing PDFP and PDFA (0.89±0.03) are significantly (p-value < 0.001) higher than those comparing PDFS and PDFA (0.83±0.05). Conclusion: Probability-driven sorting will generate multi-cycle 4D-MRI images with embedded information regarding breathing variation. Tumor motion PDFs determined in this manner are reproducible, which is important for accurate dose delivery.

Related Papers

• → Classification of Vegetable Oils by Principal Component Analysis of FTIR Spectra(2003)57 cited
• → Use of Sparse Principal Component Analysis (SPCA) for Fault Detection(2016)15 cited
• → Principal Component Analysis Versus Factor Analysis(2020)3 cited
• → Fuzzy Clustering based Principal Component Analysis(2006)3 cited
• → Principal Component Analysis versus Factor Analysis(2021)