Results for “AAPM grand challenge”

DOE OSTI2024

Report on the AAPM grand challenge on deep generative modeling for learning medical image statistics

Abstract Background The findings of the 2023 AAPM Grand Challenge on Deep Generative Modeling for Learning Medical Image Statistics are reported in this Special Report. Purpose The goal of this challenge was to promote the development of deep generative models for medical imaging and to emphasize the need for their domain‐relevant assessments via the analysis of relevant image statistics. Methods As part of this Grand Challenge, a common training dataset and an evaluation procedure was developed for benchmarking deep generative models for medical image synthesis. To create the training dataset, an established 3D virtual breast phantom was adapted. The resulting dataset comprised about 108 000 images of size 512 512. For the evaluation of submissions to the Challenge, an ensemble of 10 000 DGM‐generated images from each submission was employed. The evaluation procedure consisted of two stages. In the first stage, a preliminary check for memorization and image quality (via the Fréchet Inception Distance [FID]) was performed. Submissions that passed the first stage were then evaluated for the reproducibility of image statistics corresponding to several feature families including texture, morphology, image moments, fractal statistics, and skeleton statistics. A summary measure in this feature space was employed to rank the submissions. Additional analyses of submissions was performed to assess DGM performance specific to individual feature families, the four classes in the training data, and also to identify various artifacts. Results Fifty‐eight submissions from 12 unique users were received for this Challenge. Out of these 12 submissions, 9 submissions passed the first stage of evaluation and were eligible for ranking. The top‐ranked submission employed a conditional latent diffusion model, whereas the joint runners‐up employed a generative adversarial network, followed by another network for image superresolution. In general, we observed that the overall ranking of the top 9 submissions according to our evaluation method (i) did not match the FID‐based ranking, and (ii) differed with respect to individual feature families. Another important finding from our additional analyses was that different DGMs demonstrated similar kinds of artifacts. Conclusions This Grand Challenge highlighted the need for domain‐specific evaluation to further DGM design as well as deployment. It also demonstrated that the specification of a DGM may differ depending on its intended use.

Radiology, Nuclear Medicine & Medical Imaging↗

DOE OSTI2025

AAPM Truth‐based CT (TrueCT) reconstruction grand challenge

Background: This Special Report summarizes the 2022, AAPM grand challenge on Truth-based CT image reconstruction. Purpose: To provide an objective framework for evaluating CT reconstruction methods using virtual imaging resources consisting of a library of simulated CT projection images of a population of human models with various diseases. Methods: Two hundred unique anthropomorphic, computational models were created with varied diseases consisting of 67 emphysema, 67 lung lesions, and 66 liver lesions. The organs were modeled based on clinical CT images of real patients. The emphysematous regions were modeled using segmentations from patient CT cases in the COPDGene Phase I dataset. For the lung and liver lesion cases, 1–6 malignant lesions were created and inserted into the human models, with lesion diameters ranging from 5.6 to 21.9 mm for lung lesions and 3.9 to 14.9 mm for liver lesions. The contrast defined between the liver lesions and liver parenchyma was 82 ± 12 HU, ranging from 50 to 110 HU. Similarly, the contrast between the lung lesions and the lung parenchyma was defined as 781 ± 11 HU, ranging from 725 to 805 HU. For the emphysematous regions, the defined HU values were −950 ± 17 HU ranging from −918 to −979 HU. The developed human models were imaged with a validated CT simulator. The resulting CT sinograms were shared with the participants. The participants reconstructed CT images from the sinograms and sent back their reconstructed images. Further, the reconstructed images were then scored by comparing the results against the corresponding ground truth values. The scores included both task-generic (root mean square error [RMSE] and structural similarity matrix [SSIM]), and task-specific (detectability index [d’] and lesion volume accuracy) metrics. For the cases with multiple lesions, the measured metric was averaged across all the lesions. To combine the metrics with each other, each metric was normalized to a range of 0 to 1 per disease type, with “0” and “1” being the worst and best measured values across all cases of the disease type for all received reconstructions. Results: The True-CT challenge attracted 52 participants, out of which 5 successfully completed the challenge and submitted the requested 200 reconstructions. Across all participants and disease types, SSIM absolute values ranged from 0.22 to 0.90, RMSE from 77.6 to 490.5 HU, d’ from 0.1 to 64.6, and volume accuracy ranged from 1.2 to 753.1 mm3. The overall scores demonstrated that participant “A” had the best performance in all categories, except for the metrics of d’ for lung lesions and RMSE for liver lesions. Participant “A” had an average normalized score of 0.41 ± 0.22, 0.48 ± 0.32, and 0.42 ± 0.33 for the emphysema, lung lesion, and liver lesion cases, respectively. Conclusions: The True-CT challenge successfully enabled objective assessment of CT reconstructions with the unique advantage of access to a diverse population of diseased human models with known ground truth. This study highlights the significant potential of virtual imaging trials in objective assessment of medical imaging technologies.

60 APPLIED LIFE SCIENCES↗