Unsupervised Deep Learning Solves Computed Tomography Inverse Problems

Researchers have developed a novel unsupervised deep learning framework for solving inverse problems in computed tomography, particularly when ground-truth data is unavailable. This method, termed "Deep Image Prior" (DIP), leverages similarities between iterative reconstruction and unrolled optimization to train a network that can reconstruct unseen scans with a single forward pass. Evaluations on the 2DeteCT dataset show competitive or superior results compared to traditional methods like filtered back-projection and maximum-likelihood reconstruction, while offering a significant speed-up of approximately four orders of magnitude over per-instance DIP baselines. AI

IMPACT This unsupervised learning approach could accelerate time-critical medical imaging applications by enabling rapid reconstruction of scans without ground-truth data.

RANK_REASON This is a research paper detailing a new unsupervised deep learning method for computed tomography. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• 2DeteCT
• computed tomography
• Deep Image Prior
• Filtered back projection
• Laura Hellwege
• Maximum likelihood reconstruction for emission tomography

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →