Detection transformers speed microstructure MRI analysis and improve accuracy

Detection Instead of Inversion: T1-T2 Microstructure Quantification With Detection Transformers

Computer Vision and Pattern RecognitionMachine Learning

Summary

Measuring tiny structures inside the body with MRI is hard because the usual calculations take a long time and can be unstable. The authors turned this problem into something like finding objects in a photo by using a special AI called a detection transformer. This AI looks at the MRI signals and quickly figures out what types of tiny compartments are inside each tiny part of the body. Their new method is much faster and more accurate than traditional ways at identifying these compartments.

What this means in practice

Authors

Sebastian Endt, Fatih Özkan, Marion I. Menzel

Abstract

T1-T2 correlation MRI has strong potential to quantify microstructural compartments without imposing a biological model. Traditionally, full T1-T2 spectra per voxel are reconstructed with an ill-conditioned and time-consuming Inverse Laplace Transform. We propose to view this challenge as an object detection task: A Detection Transformer (DETR) takes 64 TI/TE signals and directly predicts a set of compartments specified by T1, T2, and signal fraction. We achieve a mean average precision mAP@10 of 76.8% on simulated data, outperforming both our previous model design (53.8%) and regularized NNLS (66.1%). We show that our DETR approach outperforms NNLS in almost every setting across compartment count, compartmental signal fraction, and SNR. Reconstruction is orders of magnitude faster than NNLS. The results are fully reproducible from our public code: github.com/SebastianEndtTHI/t1t2-detr