JSJoshua Schaefferkoetter

Joshua Schaefferkoetter

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Joshua Schaefferkoetter has a total of 45 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Deep Learning for PET Detector Crystal Segmentation

    by Baarbod Ashenagar, Johannes Breuer, Mehmet Aykac, Josh Schaefferkoetter

    Accurate definition of the crystal region map for scintillation block detectors is needed to correctly identify which crystals are associated with particular regions of the detector position profile. This task has conventionally been performed using analytical or semi-analytical methods applied to the position profiles. However, the performance of these methods relies on the quality of the position profile, and furthermore it has not been demonstrated whether these approaches are robust to sparse-count data, which is relevant for determining how long the scan time needs to be for PET scanner calibration. Here, we investigate an approach using convolutional neural networks trained using labels generated by a semi-analytical method. We show that the network approach outperformed the semi-analytical approach on sparse-count data, as well as when the position profile was substantially distorted. Our results demonstrate the potential for deep learning approaches to compliment and expand the capabilities of analytically-based crystal region map generation methods by incorporating distinct data perturbations during training. This could open new avenues for significantly reducing PET calibration scan times, realizing new detector designs, and extending the performance lifetime of installed detectors.

  • Improved Quantification in End-to-End Deep Learning FastPET Reconstruction Using Multiview Histo-Images of Attenuation Correction Factors

    by Maël Millardet, Deepak Bharkhada, Juhi Raj, Josh Schaefferkoetter, Vladimir Panin, Maurizio Conti, Samuel Matej

    End-to-end deep learning PET reconstruction significantly surpasses traditional iterative methods in speed and shows promise for surpassing them in specific scenarios, such as low-dose imaging. In 2019, a significant advancement was made by using histo-images instead of TOF sinograms as the networks input. Histo-images, by leveraging the images geometry, are more compatible with convolutional neural networks than TOF sinograms. Typically, the networks input comprises a PET data histo-image patch and an attenuation map patch. However, this method has shown inconsistent bias in the reconstructed images. This work demonstrates that bias present in the prior method can be mitigated with alternative representations of attenuation information. Instead of using the attenuation map directly, we propose using a multi-view histo-image of the attenuation correction factors, inspired by the iterative DIRECT framework and standard statistical modeling practices. We tested using them as separate channels, as well as using them to pre-correct the data, or both. This histo-image encompasses the attenuation properties of each voxel from all directions within the entire lines of response. Our approaches significantly enhances image quantification, reducing the relative difference from MLEM to an average of 2.0% to 3.0% across 16 regions of interest, compared to 9.1% with the previous method. Our statistical hypothesis test showed that the proposed methods significantly reduced absolute bias compared to the previous method, with p-values ranging from 0.002 to 0.007.


  • Open Access

    Lung lesion detectability on images obtained from decimated and CNN-based denoised [18F]-FDG PET/CT scan: an observer-based study for lung-cancer screening

    by Daphné Faist, Silvano Gnesin, Siria Medici, Alysée Khan, Marie Nicod Lalonde, Niklaus Schaefer, Adrien Depeursinge, Maurizio Conti, Joshua Schaefferkoetter, John O. Prior, et al.

    Purpose: To assess feasibility of lung cancer screening, we analysed lung lesion detectability simulating low-dose and convolutional neural network (CNN) denoised [18F]-FDG PET/CT reconstructions. Methods: Retrospectively, we analysed lung lesions on full statistics and decimated [18F]-FDG PET/CT. Reduced count PET data were emulated according to various percentage levels of total. Full and reduced statistics datasets were denoised using a CNN algorithm trained to recreate full statistics PET. Two readers assessed a detectability score from 3 to 0 for each lesion. The resulting detectability score and quantitative measurements were compared between full statistics and the different decimation levels (100%, 30%, 5%, 2%, 1%) with and without denoising. Results: We analysed 141 lung lesions from 49 patients across 588 reconstructions. The dichotomised lung lesion malignancy score was significantly different from 10% decimation without denoising (p < 0.029) and from 5% decimation with denoising (p < 0.001). Compared to full statistics, detectability score distribution differed significantly from 2% decimation without denoising (p < 0.001) and from 5% decimation with denoising (p < 0.001). Detectability scores at same decimation levels with or without denoising differed significantly at 10%, 2%, and 1% decimation (p < 0.019); dichotomised scores did not differ significantly. Denoising significantly increased the proportion of lung lesion scores with a high diagnostic confidence (3 and 0) (p < 0.038). Conclusion: Lung lesion detectability was preserved down to 30% of injected activity without denoising and to 10% with denoising. These results support the feasibility of reduced-activity [18F]-FDG PET/CT as a potential tool for lung lesion detection. Further studies are warranted to compare this approach with low-dose CT in screening settings.


  • Validation and clinical impact of motion-free PET imaging using data-driven respiratory gating and elastic PET-CT registration

    by André H. Dias, Joshua Schaefferkoetter, Josefine R. Madsen, Trine Ø. Barkholt, Mikkel H. Vendelbo, Anders B. Rodell, Noah Birge, Paul Schleyer, Ole L. Munk

    Purpose Clinical whole-body (WB) PET images can be compensated for respiratory motion using data-driven gating (DDG). However, PET DDG images may still exhibit motion artefacts at the diaphragm if the CT is acquired in a different respiratory phase than the PET image. This study evaluates the combined use of PET DDG and a deep-learning model (AIR-PETCT) for elastic registration of CT (WarpCT) to the non attenuation- and non scatter-corrected PET image (PET NAC), enabling improved PET reconstruction. Methods The validation cohort included 20 patients referred for clinical FDG PET/CT, undergoing two CT scans: a free respiration CTfree and an end-expiration breath-hold CTex. AIR-PETCT registered each CT to the PET NAC and PET DDG NAC images. The image quality of PET and PET DDG images reconstructed using CTs and WarpCTs was evaluated by three blinded readers. Additionally, a clinical impact cohort of 20 patients with significant “banana” artefacts from FDG, PSMA, and DOTATOC scans was assessed for image quality and tumor-to-background ratios. Results AIR-PETCT was robust and generated consistent WarpCTs when registering different CTs to the same PET NAC. The use of WarpCT instead of CT consistently led to equivalent or improved PET image quality. The algorithm significantly reduced “banana” artefacts and improved lesion-to-background ratios around the diaphragm. The blinded clinicians clearly preferred PET DDG images reconstructed using WarpCT. Conclusion AIR-PETCT effectively reduces respiratory motion artefacts from PET images, while improving lesion contrast. The combination of PET DDG and WarpCT holds promise for clinical application, improving PET image evaluation and diagnostic confidence.


  • Deep learning for improving PET/CT attenuation correction by elastic registration of anatomical data

    by Joshua Schaefferkoetter, Vijay Shah, Charles Hayden, John O. Prior, Sven Zuehlsdorff

    Background For PET/CT, the CT transmission data are used to correct the PET emission data for attenuation. However, subject motion between the consecutive scans can cause problems for the PET reconstruction. A method to match the CT to the PET would reduce resulting artifacts in the reconstructed images. Purpose This work presents a deep learning technique for inter-modality, elastic registration of PET/CT images for improving PET attenuation correction (AC). The feasibility of the technique is demonstrated for two applications: general whole-body (WB) imaging and cardiac myocardial perfusion imaging (MPI), with a specific focus on respiratory and gross voluntary motion. Materials and methods A convolutional neural network (CNN) was developed and trained for the registration task, comprising two distinct modules: a feature extractor and a displacement vector field (DVF) regressor. It took as input a non-attenuation-corrected PET/CT image pair and returned the relative DVF between them—it was trained in a supervised fashion using simulated inter-image motion. The 3D motion fields produced by the network were used to resample the CT image volumes, elastically warping them to spatially match the corresponding PET distributions. Performance of the algorithm was evaluated in different independent sets of WB clinical subject data: for recovering deliberate misregistrations imposed in motion-free PET/CT pairs and for improving reconstruction artifacts in cases with actual subject motion. The efficacy of this technique is also demonstrated for improving PET AC in cardiac MPI applications. Results A single registration network was found to be capable of handling a variety of PET tracers. It demonstrated state-of-the-art performance in the PET/CT registration task and was able to significantly reduce the effects of simulated motion imposed in motion-free, clinical data. Registering the CT to the PET distribution was also found to reduce various types of AC artifacts in the reconstructed PET images of subjects with actual motion. In particular, liver uniformity was improved in the subjects with significant observable respiratory motion. For MPI, the proposed approach yielded advantages for correcting artifacts in myocardial activity quantification and potentially for reducing the rate of the associated diagnostic errors. Conclusion This study demonstrated the feasibility of using deep learning for registering the anatomical image to improve AC in clinical PET/CT reconstruction. Most notably, this improved common respiratory artifacts occurring near the lung/liver border, misalignment artifacts due to gross voluntary motion, and quantification errors in cardiac PET imaging. Graphical Abstract


  • Deep learning for whole-body medical image generation

    by Joshua Schaefferkoetter, Jianhua Yan, Sangkyu Moon, Rosanna Chan, Claudia Ortega, Ur Metser, Alejandro Berlin, Patrick Veit-Haibach

    Background Artificial intelligence (AI) algorithms based on deep convolutional networks have demonstrated remarkable success for image transformation tasks. State-of-the-art results have been achieved by generative adversarial networks (GANs) and training approaches which do not require paired data. Recently, these techniques have been applied in the medical field for cross-domain image translation. Purpose This study investigated deep learning transformation in medical imaging. It was motivated to identify generalizable methods which would satisfy the simultaneous requirements of quality and anatomical accuracy across the entire human body. Specifically, whole-body MR patient data acquired on a PET/MR system were used to generate synthetic CT image volumes. The capacity of these synthetic CT data for use in PET attenuation correction (AC) was evaluated and compared to current MR-based attenuation correction (MR-AC) methods, which typically use multiphase Dixon sequences to segment various tissue types. Materials and methods This work aimed to investigate the technical performance of a GAN system for general MR-to-CT volumetric transformation and to evaluate the performance of the generated images for PET AC. A dataset comprising matched, same-day PET/MR and PET/CT patient scans was used for validation. Results A combination of training techniques was used to produce synthetic images which were of high-quality and anatomically accurate. Higher correlation was found between the values of mu maps calculated directly from CT data and those derived from the synthetic CT images than those from the default segmented Dixon approach. Over the entire body, the total amounts of reconstructed PET activities were similar between the two MR-AC methods, but the synthetic CT method yielded higher accuracy for quantifying the tracer uptake in specific regions. Conclusion The findings reported here demonstrate the feasibility of this technique and its potential to improve certain aspects of attenuation correction for PET/MR systems. Moreover, this work may have larger implications for establishing generalized methods for inter-modality, whole-body transformation in medical imaging. Unsupervised deep learning techniques can produce high-quality synthetic images, but additional constraints may be needed to maintain medical integrity in the generated data. Graphical abstract


  • Open Access

    Convolutional neural networks for improving image quality with noisy PET data

    by Josh Schaefferkoetter, Jianhua Yan, Claudia Ortega, Andrew Sertic, Eli Lechtman, Yael Eshet, Ur Metser, Patrick Veit-Haibach

    Goal: PET is a relatively noisy process compared to other imaging modalities, and sparsity of acquisition data leads to noise in the images. Recent work has focused on machine learning techniques to improve PET images, and this study investigates a deep learning approach to improve the quality of reconstructed image volumes through denoising by a 3D convolution neural network. Potential improvements were evaluated within a clinical context by physician performance in a reading task.Methods: A wide range of controlled noise levels was emulated from a set of chest PET data in patients with lung cancer, and a convolutional neural network was trained to denoise the reconstructed images using the full-count reconstructions as the ground truth. The benefits, over conventional Gaussian smoothing, were quantified across all noise levels by observer performance in an image ranking and lesion detection task.Results: The CNN-denoised images were generally ranked by the physicians equal to or better than the Gaussian-smoothed images for all count levels, with the largest effects observed in the lowest-count image sets. For the CNN-denoised images, overall lesion contrast recovery was 60% and 90% at the 1 and 20 million count levels, respectively. Notwithstanding the reduced lesion contrast recovery in noisy data, the CNN-denoised images also yielded better lesion detectability in low count levels. For example, at 1 million true counts, the average true positive detection rate was around 40% for the CNN-denoised images and 30% for the smoothed images.Conclusion: Significant improvements were found for CNN-denoising for very noisy images, and to some degree for all noise levels. The technique presented here offered however limited benefit for detection performance for images at the count levels routinely encountered in the clinic.