This web page provides several interactive demonstrations to compliment the published manuscript Improved Quantitative Parameter Estimation for Prostate T2 Relaxometry Using Convolutional Neural Networks [Bolan et al., MAGMA 2024].
This paper explores the use of neural networks to replace conventional curve fitting for quantitative MRI applications. The results show that a convolutional neural network (CNN) gave better quantitative performance, particularly in noisy regions, and demonstrated that this is a consequence of the improved representation of the noise distribution and the inherent regularization of the convolutional architecture.
This webpage hosts three interactive demonstrations using the data in the paper to help develop intuition for how the CNNs perform. Click on the cards below to explore.
Explore the synthetic and in vivo datasets used in this paper and understand their structure.
ExploreCompare the performance of different estimation methods on synthetic and in vivo datasets.
CompareEvaluate in vivo $T_2$ maps comparing the performance of NLLS and CNN methods with varying noise levels.
EvaluateFor more information see the original project repository, read the final paper, or the preprint.
For an quick overview, watch the demo video on the right (6 min). This was for a digital poster presented at the 2023 annual meeting of the International Society for Magnetic Resonance (ISMRM).