Search

Diffusion models demonstrate state-of-the-art performance on image generation, and are gaining traction for sparse medical image reconstruction tasks. However, compared to classical reconstruction algorithms relying on simple analytical priors, diffusion models have the dangerous property of producing realistic looking results \emph{even when incorrect}, particularly with few observations. We investigate the utility of diffusion models as priors for image reconstruction by varying the number of observations and comparing their performance to classical priors (sparse and Tikhonov regularization) using pixel-based, structural, and downstream metrics. We make comparisons on low-dose chest wall computed tomography (CT) for fat mass quantification. First, we find that classical priors are superior to diffusion priors when the number of projections is ``sufficient''. Second, we find that diffusion priors can capture a large amount of detail with very few observations, significantly outperforming classical priors. However, they fall short of capturing all details, even with many observations. Finally, we find that the performance of diffusion priors plateau after extremely few ($\approx$10-15) projections. Ultimately, our work highlights potential issues with diffusion-based sparse reconstruction and underscores the importance of further investigation, particularly in high-stakes clinical settings., Comment: Accepted at IEEE ISBI 2025, 5 pages, 2 figures, 1 table

Uncertainty quantification is crucial to account for the imperfect predictions of machine learning algorithms for high-impact applications. Conformal prediction (CP) is a powerful framework for uncertainty quantification that generates calibrated prediction intervals with valid coverage. In this work, we study how CP intervals are affected by bias - the systematic deviation of a prediction from ground truth values - a phenomenon prevalent in many real-world applications. We investigate the influence of bias on interval lengths of two different types of adjustments -- symmetric adjustments, the conventional method where both sides of the interval are adjusted equally, and asymmetric adjustments, a more flexible method where the interval can be adjusted unequally in positive or negative directions. We present theoretical and empirical analyses characterizing how symmetric and asymmetric adjustments impact the "tightness" of CP intervals for regression tasks. Specifically for absolute residual and quantile-based non-conformity scores, we prove: 1) the upper bound of symmetrically adjusted interval lengths increases by $2|b|$ where $b$ is a globally applied scalar value representing bias, 2) asymmetrically adjusted interval lengths are not affected by bias, and 3) conditions when asymmetrically adjusted interval lengths are guaranteed to be smaller than symmetric ones. Our analyses suggest that even if predictions exhibit significant drift from ground truth values, asymmetrically adjusted intervals are still able to maintain the same tightness and validity of intervals as if the drift had never happened, while symmetric ones significantly inflate the lengths. We demonstrate our theoretical results with two real-world prediction tasks: sparse-view computed tomography (CT) reconstruction and time-series weather forecasting. Our work paves the way for more bias-robust machine learning systems., Comment: 17 pages, 6 figures, code available at: https://github.com/matthewyccheung/conformal-metric

Computed Tomography (CT) scans are the standard-of-care for the visualization and diagnosis of many clinical ailments, and are needed for the treatment planning of external beam radiotherapy. Unfortunately, the availability of CT scanners in low- and mid-resource settings is highly variable. Planar x-ray radiography units, in comparison, are far more prevalent, but can only provide limited 2D observations of the 3D anatomy. In this work we propose DIFR3CT, a 3D latent diffusion model, that can generate a distribution of plausible CT volumes from one or few (<10) planar x-ray observations. DIFR3CT works by fusing 2D features from each x-ray into a joint 3D space, and performing diffusion conditioned on these fused features in a low-dimensional latent space. We conduct extensive experiments demonstrating that DIFR3CT is better than recent sparse CT reconstruction baselines in terms of standard pixel-level (PSNR, SSIM) on both the public LIDC and in-house post-mastectomy CT datasets. We also show that DIFR3CT supports uncertainty quantification via Monte Carlo sampling, which provides an opportunity to measure reconstruction reliability. Finally, we perform a preliminary pilot study evaluating DIFR3CT for automated breast radiotherapy contouring and planning -- and demonstrate promising feasibility. Our code is available at https://github.com/yransun/DIFR3CT., Comment: 11 pages, 9 figures

Modern deep learning reconstruction algorithms generate impressively realistic scans from sparse inputs, but can often produce significant inaccuracies. This makes it difficult to provide statistically guaranteed claims about the true state of a subject from scans reconstructed by these algorithms. In this study, we propose a framework for computing provably valid prediction bounds on claims derived from probabilistic black-box image reconstruction algorithms. The key insights behind our framework are to represent reconstructed scans with a derived clinical metric of interest, and to calibrate bounds on the ground truth metric with conformal prediction (CP) using a prior calibration dataset. These bounds convey interpretable feedback about the subject's state, and can also be used to retrieve nearest-neighbor reconstructed scans for visual inspection. We demonstrate the utility of this framework on sparse-view computed tomography (CT) for fat mass quantification and radiotherapy planning tasks. Results show that our framework produces bounds with better semantical interpretation than conventional pixel-based bounding approaches. Furthermore, we can flag dangerous outlier reconstructions that look plausible but have statistically unlikely metric values., Comment: 11 pages, 4 figures, 1 table, 2 algorithms. Code available at https://github.com/matthewyccheung/conformal-metric. Previously titled "Metric-guided Image Reconstruction Bounds via Conformal Prediction"

Deep learning has significantly advanced the potential for automated contouring in radiotherapy planning. In this manuscript, guided by contemporary literature, we underscore three key insights: (1) High-quality training data is essential for auto-contouring algorithms; (2) Auto-contouring models demonstrate commendable performance even with limited medical image data; (3) The quantitative performance of auto-contouring is reaching a plateau. Given these insights, we emphasize the need for the radiotherapy research community to embrace data-centric approaches to further foster clinical adoption of auto-contouring technologies., Comment: 13 pages, 4 figures

CT scans are the standard-of-care for many clinical ailments, and are needed for treatments like external beam radiotherapy. Unfortunately, CT scanners are rare in low and mid-resource settings due to their costs. Planar X-ray radiography units, in comparison, are far more prevalent, but can only provide limited 2D observations of the 3D anatomy. In this work, we propose a method to generate CT volumes from few (<5) planar X-ray observations using a prior data distribution, and perform the first evaluation of such a reconstruction algorithm for a clinical application: radiotherapy planning. We propose a deep generative model, building on advances in neural implicit representations to synthesize volumetric CT scans from few input planar X-ray images at different angles. To focus the generation task on clinically-relevant features, our model can also leverage anatomical guidance during training (via segmentation masks). We generated 2-field opposed, palliative radiotherapy plans on thoracic CTs reconstructed by our method, and found that isocenter radiation dose on reconstructed scans have <1% error with respect to the dose calculated on clinically acquired CTs using <=4 X-ray views. In addition, our method is better than recent sparse CT reconstruction baselines in terms of standard pixel and structure-level metrics (PSNR, SSIM, Dice score) on the public LIDC lung CT dataset. Code is available at: https://github.com/wanderinrain/Xray2CT., Comment: 10 pages, 5 figures

Purpose: Target delineation for radiation therapy is a time-consuming and complex task. Autocontouring gross tumor volumes (GTVs) has been shown to increase efficiency. However, there is limited literature on post-operative target delineation, particularly for CT-based studies. To this end, we trained a CT-based autocontouring model to contour the post-operative GTV of pediatric patients with medulloblastoma. Methods: 104 retrospective pediatric CT scans were used to train a GTV auto-contouring model. 80 patients were then preselected for contour visibility, continuity, and location to train an additional model. Each GTV was manually annotated with a visibility score based on the number of slices with a visible GTV (1 = <25%, 2 = 25%-50%, 3 = >50%-75%, and 4 = >75%-100%). Contrast and the contrast-to-noise ratio (CNR) were calculated for the GTV contour with respect to a cropped background image. Both models were tested on the original and pre-selected testing sets. The resulting surface and overlap metrics were calculated comparing the clinical and autocontoured GTVs and the corresponding clinical target volumes (CTVs). Results: 80 patients were pre-selected to have a continuous GTV within the posterior fossa. Of these, 7, 41, 21, and 11 were visibly scored as 4, 3, 2, and 1, respectively. The contrast and CNR removed an additional 11 and 20 patients from the dataset, respectively. The Dice similarity coefficients (DSC) were 0.61 +/- 0.29 and 0.67 +/- 0.22 on the models without pre-selected training data and 0.55 +/- 13.01 and 0.83 +/- 0.17 on the models with pre-selected data, respectively. The DSC on the CTV expansions were 0.90 +/- 0.13. Conclusion: We automatically contoured continuous GTVs within the posterior fossa on scans that had contrast >=10 HU. CT-Based auto-contouring algorithms have potential to positively impact centers with limited MRI access., Comment: 14 pages, 6 figures

Purpose: This study aimed to use deep learning-based dose prediction to assess head and neck (HN) plan quality and identify suboptimal plans. Methods: A total of 245 VMAT HN plans were created using RapidPlan knowledge-based planning (KBP). A subset of 112 high-quality plans was selected under the supervision of an HN radiation oncologist. We trained a 3D Dense Dilated U-Net architecture to predict 3-dimensional dose distributions using 3-fold cross-validation on 90 plans. Model inputs included CT images, target prescriptions, and contours for targets and organs at risk (OARs). The model's performance was assessed on the remaining 22 test plans. We then tested the application of the dose prediction model for automated review of plan quality. Dose distributions were predicted on 14 clinical plans. The predicted versus clinical OAR dose metrics were compared to flag OARs with suboptimal normal tissue sparing using a 2 Gy dose difference or 3% dose-volume threshold. OAR flags were compared to manual flags by 3 HN radiation oncologists. Results: The predicted dose distributions were of comparable quality to the KBP plans. The differences between the predicted and KBP-planned D1%, D95%, and D99% across the targets were within -2.53%(SD=1.34%), -0.42%(SD=1.27%), and -0.12%(SD=1.97%), respectively, and the OAR mean and maximum doses were within -0.33Gy(SD=1.40Gy) and -0.96Gy(SD=2.08Gy). For the plan quality assessment study, radiation oncologists flagged 47 OARs for possible plan improvement. There was high interphysician variability; 83% of physician-flagged OARs were flagged by only one of 3 physicians. The comparative dose prediction model flagged 63 OARs, including 30 of 47 physician-flagged OARs. Conclusion: Deep learning can predict high-quality dose distributions, which can be used as comparative dose distributions for automated, individualized assessment of HN plan quality., Comment: updated to reflect the published peer-reviewed article

In this work, we developed and evaluated a novel pipeline consisting of two landmark-based field aperture generation approaches for WBRT treatment planning; they are fully automated and customizable. The automation pipeline is beneficial for both clinicians and patients, where we can reduce clinician workload and reduce treatment planning time. The customizability of the field aperture design addresses different clinical requirements and allows the personalized design to become feasible. The performance results regarding quantitative and qualitative evaluations demonstrated that our plans were comparable with the original clinical plans. This technique has been deployed as part of a fully automated treatment planning tool for whole-brain cancer and could be translated to other treatment sites in the future., Comment: 10 pages, 7 figures, submitted to JACMP - Journal of Applied Clinical Medical Physics

To develop an automated workflow for rectal cancer three-dimensional conformal radiotherapy treatment planning that combines deep-learning(DL) aperture predictions and forward-planning algorithms. We designed an algorithm to automate the clinical workflow for planning with field-in-field. DL models were trained, validated, and tested on 555 patients to automatically generate aperture shapes for primary and boost fields. Network inputs were digitally reconstructed radiography, gross tumor volume(GTV), and nodal GTV. A physician scored each aperture for 20 patients on a 5-point scale(>3 acceptable). A planning algorithm was then developed to create a homogeneous dose using a combination of wedges and subfields. The algorithm iteratively identifies a hotspot volume, creates a subfield, and optimizes beam weight all without user intervention. The algorithm was tested on 20 patients using clinical apertures with different settings, and the resulting plans(4 plans/patient) were scored by a physician. The end-to-end workflow was tested and scored by a physician on 39 patients using DL-generated apertures and planning algorithms. The predicted apertures had Dice scores of 0.95, 0.94, and 0.90 for posterior-anterior, laterals, and boost fields, respectively. 100%, 95%, and 87.5% of the posterior-anterior, laterals, and boost apertures were scored as clinically acceptable, respectively. Wedged and non-wedged plans were clinically acceptable for 85% and 50% of patients, respectively. The final plans hotspot dose percentage was reduced from 121%($\pm$ 14%) to 109%($\pm$ 5%) of prescription dose. The integrated end-to-end workflow of automatically generated apertures and optimized field-in-field planning gave clinically acceptable plans for 38/39(97%) of patients. We have successfully automated the clinical workflow for generating radiotherapy plans for rectal cancer for our institution.

Vertebral labelling and segmentation are two fundamental tasks in an automated spine processing pipeline. Reliable and accurate processing of spine images is expected to benefit clinical decision-support systems for diagnosis, surgery planning, and population-based analysis on spine and bone health. However, designing automated algorithms for spine processing is challenging predominantly due to considerable variations in anatomy and acquisition protocols and due to a severe shortage of publicly available data. Addressing these limitations, the Large Scale Vertebrae Segmentation Challenge (VerSe) was organised in conjunction with the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2019 and 2020, with a call for algorithms towards labelling and segmentation of vertebrae. Two datasets containing a total of 374 multi-detector CT scans from 355 patients were prepared and 4505 vertebrae have individually been annotated at voxel-level by a human-machine hybrid algorithm (https://osf.io/nqjyw/, https://osf.io/t98fz/). A total of 25 algorithms were benchmarked on these datasets. In this work, we present the the results of this evaluation and further investigate the performance-variation at vertebra-level, scan-level, and at different fields-of-view. We also evaluate the generalisability of the approaches to an implicit domain shift in data by evaluating the top performing algorithms of one challenge iteration on data from the other iteration. The principal takeaway from VerSe: the performance of an algorithm in labelling and segmenting a spine scan hinges on its ability to correctly identify vertebrae in cases of rare anatomical variations. The content and code concerning VerSe can be accessed at: https://github.com/anjany/verse., Comment: Challenge report for the VerSe 2019 and 2020. Published in Medical Image Analysis (DOI: https://doi.org/10.1016/j.media.2021.102166)