Memory Efficient Invertible Neural Networks for 3D Photoacoustic Imaging
This addresses memory constraints in medical imaging for clinical applications like tumor monitoring, though it is incremental as it applies an existing method to a new domain.
The paper tackled the problem of memory-intensive training in machine learning methods for 3D photoacoustic imaging, which is hindered by limited-view data and artifacts, and demonstrated that invertible neural networks enable training of arbitrary depth models on a consumer GPU with 16GB RAM.
Photoacoustic imaging (PAI) can image high-resolution structures of clinical interest such as vascularity in cancerous tumor monitoring. When imaging human subjects, geometric restrictions force limited-view data retrieval causing imaging artifacts. Iterative physical model based approaches reduce artifacts but require prohibitively time consuming PDE solves. Machine learning (ML) has accelerated PAI by combining physical models and learned networks. However, the depth and overall power of ML methods is limited by memory intensive training. We propose using invertible neural networks (INNs) to alleviate memory pressure. We demonstrate INNs can image 3D photoacoustic volumes in the setting of limited-view, noisy, and subsampled data. The frugal constant memory usage of INNs enables us to train an arbitrary depth of learned layers on a consumer GPU with 16GB RAM.