Generation of virtual lung single‐photon emission computed tomography/CT fusion images for functional avoidance radiotherapy planning using machine learning algorithms

Jang, Bong Seok; Chang, Ji Hyun; Park, Andrew J.; Wu, Hong Gyun

doi:10.1111/1754-9485.12868

Cited by 6 publications

(5 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As their ground truth, Zhong et al used a synthetic image which was computationally derived from 4DCT using an image registration technique (Castillo et al 2019). Jang et al (2019) generated SPECT perfusion from CT images with a 2D-UNet (Ronneberger et al 2015) based cGAN (Isola et al 2017) model trained on the inhalation CT acquired during the SPECT/CT scan. As the authors note, the 2D model design presented inherent performance limitations, notably making the predicted volumes prone to discontinuities.…”

Section: Introductionmentioning

confidence: 99%

Synthetic pulmonary perfusion images from 4DCT for functional avoidance using deep learning

et al. 2021

View full text Add to dashboard Cite

Purpose. To develop and evaluate the performance of a deep learning model to generate synthetic pulmonary perfusion images from clinical 4DCT images for patients undergoing radiotherapy for lung cancer. Methods. A clinical data set of 58 pre- and post-radiotherapy 99mTc-labeled MAA-SPECT perfusion studies (32 patients) each with contemporaneous 4DCT studies was collected. Using the inhale and exhale phases of the 4DCT, a 3D-residual network was trained to create synthetic perfusion images utilizing the MAA-SPECT as ground truth. The training process was repeated for a 50-imaging study, five-fold validation with twenty model instances trained per fold. The highest performing model instance from each fold was selected for inference upon the eight-study test set. A manual lung segmentation was used to compute correlation metrics constrained to the voxels within the lungs. From the pre-treatment test cases (N = 5), 50th percentile contours of well-perfused lung were generated from both the clinical and synthetic perfusion images and the agreement was quantified. Results. Across the hold-out test set, our deep learning model predicted perfusion with a Spearman correlation coefficient of 0.70 (IQR: 0.61–0.76) and a Pearson correlation coefficient of 0.66 (IQR: 0.49–0.73). The agreement of the functional avoidance contour pairs was Dice of 0.803 (IQR: 0.750–0.810) and average surface distance of 5.92 mm (IQR: 5.68–7.55). Conclusion. We demonstrate that from 4DCT alone, a deep learning model can generate synthetic perfusion images with potential application in functional avoidance treatment planning.

show abstract

Section: Introductionmentioning

confidence: 99%

Synthetic pulmonary perfusion images from 4DCT for functional avoidance using deep learning

et al. 2021

View full text Add to dashboard Cite

show abstract

“…GANs have continued to show promise in synthesis problems. 119 CT images have been used to generate SPECT images via a conditional GAN (cGAN) instead of a CNN. 29 The method used a 2D GAN with 49 patients consisting of 3054 2D images as training data; the testing data contains 5 patients.…”

Section: Resultsmentioning

confidence: 99%

“…A key limitation for synthesis methods is the errors introduced by the registration of source and target images. Consequently, it has been suggested that images that are not matched anatomically due to breathing discrepancies are excluded, 119 complicating validation for clinical adoption. 29,119 …”

Section: Resultsmentioning

confidence: 99%

Deep learning in structural and functional lung image analysis

Wild

Tahir

2021

The British Journal of Radiology

View full text Add to dashboard Cite

The recent resurgence of deep learning (DL) has dramatically influenced the medical imaging field. Medical image analysis applications have been at the forefront of DL research efforts applied to multiple diseases and organs, including those of the lungs. The aims of this review are twofold: (i) to briefly overview DL theory as it relates to lung image analysis; (ii) to systematically review the DL research literature relating to the lung image analysis applications of segmentation, reconstruction, registration and synthesis. The review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. 479 studies were initially identified from the literature search with 82 studies meeting the eligibility criteria. Segmentation was the most common lung image analysis DL application (65.9% of papers reviewed). DL has shown impressive results when applied to segmentation of the whole lung and other pulmonary structures. DL has also shown great potential for applications in image registration, reconstruction and synthesis. However, the majority of published studies have been limited to structural lung imaging with only 12.9% of reviewed studies employing functional lung imaging modalities, thus highlighting significant opportunities for further research in this field. Although the field of DL in lung image analysis is rapidly expanding, concerns over inconsistent validation and evaluation strategies, intersite generalisability, transparency of methodological detail and interpretability need to be addressed before widespread adoption in clinical lung imaging workflow.

show abstract

“…Additionally, in many articles the normalization strategy is not mentioned and, hence, it is difficult to ascertain the risk of bias. 1,4,6,[29][30][31][32][33] Finally, some obstacles to the interarticle comparisons of SSIM are identified, notably the hyperparameters influencing the SSIM computation such as the dynamic range, the type of evaluation (two TA B L E 1 Scoping review illustrating potentially problematic SSIM scenarios, such as using SSIM on HU (SSIM on the CT range), using SSIM on normalized images potentially containing negative numbers (normalization) and evaluating SSIM on images without mention of the applied normalization (unclear normalization)…”

Section: Related Workmentioning

confidence: 99%

“…In multiple settings, SSIM was used as a cost function 26–28 in images that might have contained negative values. Additionally, in many articles the normalization strategy is not mentioned and, hence, it is difficult to ascertain the risk of bias 1,4,6,29–33 . Table 1 summarizes the nonexhaustive literature review from the previous section.…”

Section: Introductionmentioning

confidence: 99%

On the proper use of structural similarity for the robust evaluation of medical image synthesis models

2022

View full text Add to dashboard Cite

To propose good practices for using the structural similarity metric (SSIM) and reporting its value. SSIM is one of the most popular image quality metrics in use in the medical image synthesis community because of its alleged superiority over voxel-by-voxel measurements like the average error or the peak signal noise ratio (PSNR). It has seen massive adoption since its introduction, but its limitations are often overlooked. Notably, SSIM is designed to work on a strictly positive intensity scale, which is generally not the case in medical imaging. Common intensity scales such as the Houndsfield units (HU) contain negative numbers, and they can also be introduced by image normalization techniques such as the z-normalization. Methods: We created a series of experiments to quantify the impact of negative values in the SSIM computation. Specifically, we trained a three-dimensional (3D) U-Net to synthesize T2-weighted MRI from T1-weighted MRI using the BRATS 2018 dataset. SSIM was computed on the synthetic images with a shifted dynamic range. Next, to evaluate the suitability of SSIM as a loss function on images with negative values, it was used as a loss function to synthesize znormalized images. Finally, the difference between two-dimensional (2D) SSIM and 3D SSIM was investigated using multiple 2D U-Nets trained on different planes of the images. Results:The impact of the misuse of the SSIM was quantified; it was established that it introduces a large downward bias in the computed SSIM. It also introduces a small random error that can change the relative ranking of models. The exact values for this bias and error depend on the quality and the intensity histogram of the synthetic images. Although small, the reported error is significant considering the small SSIM difference between state-of -the-art models. It was shown therefore that SSIM cannot be used as a loss function when images contain negative values due to major errors in the gradient calculation, resulting in under-performing models. 2D SSIM was also found to be overestimated in 2D image synthesis models when computed along the plane of synthesis, due to the discontinuities between slices that is typical of 2D synthesis methods. Conclusion: Various types of misuse of the SSIM were identified, and their impact was quantified. Based on the findings, this paper proposes good practices when using SSIM, such as reporting the average over the volume of the image containing tissue and appropriately defining the dynamic range.

show abstract

Generation of virtual lung single‐photon emission computed tomography/CT fusion images for functional avoidance radiotherapy planning using machine learning algorithms

Cited by 6 publications

References 29 publications

Synthetic pulmonary perfusion images from 4DCT for functional avoidance using deep learning

Synthetic pulmonary perfusion images from 4DCT for functional avoidance using deep learning

Deep learning in structural and functional lung image analysis

On the proper use of structural similarity for the robust evaluation of medical image synthesis models

Contact Info

Product

Resources

About