SAVE-Net: Deep Learning Synthesized DSA Sequences for Radiation Dose Reduction

A multicenter study presents SAVE-Net, a deep learning model for synthesizing digital subtraction angiography (DSA) frames. The model was trained on 17,335 DSA sequences from one hospital and externally validated on 3,255 sequences from two other hospitals. Results indicate the model can generate diagnostic sequences using 1/7 of the standard radiation dose, with image quality comparable to real data.

Use Cases

• Training deep learning models for medical image synthesis based on DSA sequences.
• Evaluating image quality metrics like SSIM and PSNR for generated medical frames.
• Conducting visual Turing tests and quality assessments with clinical experts.
• Developing protocols for radiation dose reduction in cerebrovascular imaging procedures.

Strengths

• Multicenter validation involving 3,255 external DSA sequences from two hospitals.
• Quantitative performance metrics reported: SSIM of 0.951 and PSNR of 40.764.
• Model generation speed is 0.04 seconds per frame.
• Clinical assessment by five interventional radiologists showed no significant difference from real data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic or institutional bias inherent to the three source hospitals.

Provenance

Source

Ruibo Liu via figshare

Collection Method

DSA sequences collected from three hospitals for model training and validation.

Freshness

Last updated 2026-05-15 05:41:07; freshness should be verified.

Geography

Data sourced from three hospitals; specific locations unknown.