Clinical evaluation of deep learning-enhanced lymphoma pet imaging with accelerated acquisition

Abstract

Purpose: This study aims to evaluate the clinical performance of a deep learning (DL)-enhanced two-fold accelerated PET imaging method in patients with lymphoma.

Methods: A total of 123 cases devoid of lymphoma underwent whole-body 18F-FDG-PET/CT scans to facilitate the development of an advanced SAU2Net model, which combines the advantages of U2Net and attention mechanism. This model integrated inputs from simulated 1/2-dose (0.07 mCi/kg) PET acquisition across multiple slices to generate an estimated standard dose (0.14 mCi/kg) PET scan. Additional 39 cases with confirmed lymphoma pathology were utilized to evaluate the model's clinical performance. Assessment criteria encompassed peak-signal-to-noise ratio (PSNR), structural similarity index (SSIM), a 5-point Likert scale rated by two experienced physicians, SUV features, image noise in the liver, and contrast-to-noise ratio (CNR). Diagnostic outcomes, including lesion numbers and Deauville score, were also compared.

Results: Images enhanced by the proposed DL method exhibited superior image quality (P < 0.001) in comparison to low-dose acquisition. Moreover, they illustrated equivalent image quality in terms of subjective image analysis and lesion maximum standardized uptake value (SUVmax) as compared to the standard acquisition method. A linear regression model with y = 1.017x + 0.110 ( $R^2=1.00$ ) can be established between the enhanced scans and the standard acquisition for lesion SUVmax. With enhancement, increased signal-to-noise ratio (SNR), CNR, and reduced image noise were observed, surpassing those of the standard acquisition. DL-enhanced PET images got diagnostic results essentially equavalent to standard PET images according to two experienced readers.

Conclusion: The proposed DL method could facilitate a 50% reduction in PET imaging duration for lymphoma patients, while concurrently preserving image quality and diagnostic accuracy.

Keywords: PET; deep learning; low‐dose imaging; lymphoma.

FIGURE 1

Architecture of proposed SAU2Net. The network backbone is a U2Net model with batch normalization layers removed. Adapted CBAM blocks are incorporated within the long skip connections.

FIGURE 2

The adapted CBAM module consists of a channel attention module and a spatial attention module. A batch normalization layer is introduced before pooling layers to adapt the data distribution of PET images represented in SUVs.

FIGURE 3

SSIM (histogram) and PSNR (KDE curves) with enhancement method proposed. Low, DL represent low‐dose—standard dose experiment group and deep learning enhanced low‐dose—standard dose group correspondingly.

FIGURE 4

Fitted linear regressions and Pearson correlations coefficient for lesion SUV and liver SUV. Where x represents SUV value from the standard acquisition and y stands for SUV value from DL‐enhanced image. The shaded regions surrounding the lines indicate the regions of 99% confidence level SD Liver SUV: Standard deviation of counts within hepatic region which is considered as benchmark level.

FIGURE 5

Bland‐Altman analysis of SUV for lesion and liver. Where solid line in the middle and two dashed lines aside are mean difference and ± 1.96 standard deviations, region with two dashed line is with 95% limits of agreement (LOA). x and y axis are mean value and difference value each.

FIGURE 6

Boxplot of Noise, SNR and CNR on low‐dose images (Low), standard‐dose images (Standard) and enhanced images (DL) apiece.

FIGURE 7

Visualization of low‐dose images (a, d), restored images that were generated using our method on the low‐dose images (b, e) and standard‐dose images (c, f) containing maximum intensity projection (MIP) and hepatic region. The restoration process effectively elevated the visual quality of the low‐dose image to a level that was visually comparable to the standard‐dose image.

FIGURE 8

Visualization of low‐dose images (a, f), low‐dose restored images (b, g), standard‐dose images (c, h), standard‐dose restored images (d, i) and time‐added images (e, j) containing maximum intensity projection (MIP) and hepatic region. The restored process effectively elevated the visual quality of the low‐dose images to a level that was visually beyond the standard‐dose images, but still worse than that of the time‐added images. The restored process elevated the visual quality of the standard‐dose images to a level that visually comparable to the time‐added images.