Accelerating imaging: deep learning for enhanced ¹²³I-ioflupane SPECT efficiency

Area of Science:

• Nuclear Medicine
• Artificial Intelligence in Medical Imaging
• Radiopharmaceutical Imaging

Background:

• Conventional ¹²³I-ioflupane dopamine-transporter SPECT scans require 25-40 minutes, leading to patient discomfort and limited throughput.
• Deep learning (DL) reconstruction offers a potential solution to reduce scan times while maintaining image quality.

Purpose of the Study:

• To assess the feasibility of using DL reconstruction to generate diagnostic-quality ¹²³I-ioflupane SPECT images from significantly reduced 5-minute acquisition times.
• To compare the image quality and diagnostic performance of DL-reconstructed 5-minute scans with conventional 25-minute scans.

Main Methods:

• Retrospective analysis of 207 ¹²³I-ioflupane SPECT studies.
• Training and validation of six convolutional neural network architectures (U-Net variants, V-Net, Attention U-Net, TransUNet) to translate 5-minute scans into virtual 25-minute scans.
• Quantitative image quality assessment using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM).
• Blinded reader study with three nuclear medicine physicians to evaluate diagnostic performance and inter-observer agreement.

Main Results:

• All DL reconstructions significantly improved image quality (PSNR, SSIM) compared to raw 5-minute scans (p < 0.01).
• A compact four-layer U-Net achieved the highest image quality, statistically indistinguishable from 25-minute scans (p > 0.05).
• Reader concordance improved from fair (κ = 0.29-0.41) to substantial (κ = 0.62-0.70) with DL reconstruction, with high intra- and inter-observer reliability (ICC).

Conclusions:

• A four-layer U-Net deep learning model can restore diagnostic fidelity to 5-minute ¹²³I-ioflupane SPECT scans.
• This DL-accelerated protocol enables an 80% reduction in scan time without compromising quantitative metrics or diagnostic interpretability.
• DL-accelerated SPECT protocols have the potential to enhance patient comfort, reduce motion artifacts, and increase imaging throughput, warranting further prospective validation.