Reid F. Thompson

Accessible imaging technologies and stronger machine learning (ML) architectures have spurred a race to develop models to complete dermatological tasks like automated skin cancer diagnosis. However, models with high performance on benchmark datasets deteriorate when challenged with data from disparate clinical sources. Generalization gaps stem from high variability in skin lesion images from capture angle, imaging technology, and patient phenotype, among other factors, impeding the safe application of ML models. We apply a novel multi-criterion uncertainty-estimation approach to detect out-of-distribution skin lesion images from five publicly available datasets across seven countries. Using our method, Supervised Autoencoders for Generalization Estimates (SAGE), we quantify likeness of images from patients in Argentina, Brazil, Austria, North Macedonia, Turkey, Australia, and the United States to the popular HAM10000 benchmarking dataset and identify problematic image artifacts affecting the reliability of predictions in a pre-clinical setting. We show how filtering images based on SAGE score can improve the performance of a separate malignancy prediction model. We benchmark SAGE across varying severities of distribution shift (e.g., image modality and new diagnostic classes), providing users with a powerful tool for interrogating differences between their data and the training distribution of an ML model before clinical implementation.

PURPOSE: Oregon's vast and complex terrain creates potential barriers for patients needing daily radiotherapy. The large geographic area and the lack of a standardized database to guide referrals make navigating regional radiation oncology (RO) resources and understanding treatment access across the state challenging. METHODS: This is a cross-sectional statewide analysis of RO centers and physicians in Oregon using geospatial information software to estimate resident-level access, RO center–specific capability, and facility-level population coverage by distance and drive-time in 2025. Additionally, regional all-cancer outcomes are summarized by mortality-to-incidence ratios (MIRs) based on proximity to RO access. RESULTS: Oregon has 27 active RO centers, with a recent closure in Pendleton that previously served approximately 110,000 residents. Although 71% of the population lives within 10 miles of an RO facility, 54.5% of the state's land area lacks proximal access, leaving 434,855 residents (10.3%) more than 60 minutes from care. For many counties, the nearest center is located out-of-state at distances up to 137 miles. One-way drive-times vary widely but can exceed 2.5 hours. Despite statewide resource levels exceeding national averages (one linear accelerator per approximately 83,000 residents and one radiation oncologist per approximately 49,000 residents), there is limited access to brachytherapy and no proton therapy centers. Additionally, increasing geographic isolation from RO access is ecologically associated with worse all-cancer MIRs. CONCLUSION: This study provides a comprehensive characterization of RO resources across Oregon. Results reveal substantial regional disparities in proximity, treatment capability, and outcomes, particularly in rural/remote counties that lack in-state RO options. By integrating geospatial, population, and outcome data, this work establishes a statewide baseline to identify persistent access gaps and serves as both a referral resource and a framework for future access studies.

PURPOSE: Radiotherapy is a critically important cancer treatment; however, its details are often not well represented in electronic health record data sets. US Veterans' radiation courses are further distributed across a range of medical centers, both internal and external to the Veterans Health Administration (VHA), inhibiting analysis of radiotherapy treatment across this population. METHODS: We train and test a suite of supervised machine learning models for the accurate prediction of radiation course dates using billing and diagnostic codes from a combination of VHA and Centers for Medicare and Medicaid Services (CMS) databases. We use a separate heuristic algorithm to assemble course date predictions into complete radiation treatments. RESULTS: Our top model predicts radiation course dates with compelling accuracy (macro-average of 0.914 across classes). The retrospective application of our model and assembly algorithm to radiation procedure dates for 1,331,342 patients identified 1,526,660 predicted courses of radiotherapy. CONCLUSION: The identified courses were collected into a shared resource to facilitate future VHA-based studies, and our predictive model is available for application to a wider range of non-VHA data sets, particularly those leveraging CMS data.

Motivation: Cancer-specific neoepitopes may arise from abnormal splicing in the transcriptomic landscape (alternative splicing neoepitopes, ASNs), leading to divergent proteins with high potential immunogenicity. Predicting candidate ASNs requires a series of algorithmic and parameter choices, but no previous study has investigated the consistency and interpretability of these choices. Results: We apply two ASN prediction methods with 35 matched parameter sets to generate candidate ASNs for five breast and ovarian cancer samples from The Cancer Genome Atlas (TCGA). We find that: 1) junction 9-mer peptides generated by two similarly designed pipelines differed by, on average, 68.9% and 76.6% in the BRCA and OV cohorts, respectively, and putatively cancer-specific junction 9-mers from the two pipelines diverged further, by an average of 81.5 % in OV and 84.6 % in BRCA; 2) the most lenient filters in the BRCA cohort show the highest divergence, at 97 %; 3) the rate of mass spectrometry validation of ASNs’ protein presence in cells is dominated by the size of the input space; and 4) putatively cancer-specific ASNs found by the intersection of both pipelines can be validated when accounting for false discovery with an average of 1.74 and 284.4 candidates in the BRCA and OV cohorts, respectively. Taken together, these results highlight that ASN identification is fragile and difficult to reproduce across analysis platforms, with limited cross-pipeline overlap and strong dependence on parameter choices. Availability and Implementation: Python research code and scripts are available for download at https://github.com/ratschlab/projects2020_ohsu . Contact: thompsre@ohsu.edu , raetsch@inf.ethz.ch

Widespread access to imaging technologies and stronger machine learning (ML) architectures for dermatology tasks such as malignancy prediction have spurred a race to develop models to assist in the automated diagnosis of skin cancer. However, high diagnostic performance on benchmarking datasets quickly deteriorates when models are challenged with data from disparate clinical sources. Generalization gaps stem from the high variability in skin lesion images due to lighting, capture angle, imaging technology and patient phenotype among other factors, impeding the safe application of diagnostic ML models in practice. In this study, we apply a novel multi-criterion uncertainty-estimation approach to detect out-of-distribution skin lesion images from four publicly available datasets across five countries. Using our method, Supervised Autoencoders for Generalization Estimates (SAGE), we quantify likeness of images from patients in Argentina, Brazil, Austria, North Macedonia, Turkey, Australia and the United States to the popular HAM10000 benchmarking dataset and identify problematic image artifacts affecting the reliability of predictions in a pre-clinical setting. We show how filtering images based on SAGE score thresholds can improve the performance of a separate malignancy prediction model and how our approach is robust to variations in image modality and the introduction of new diagnostic classes, providing users with a powerful tool for interrogating key differences between their data and the training distribution of an ML model before clinical implementation.

Early detection of skin cancer is crucial for effective treatment and improved patient outcomes. Recent advancements in oncologic imaging, particularly molecular imaging techniques, have revolutionized cancer diagnostics and treatment by enabling the visualization of tumors and cellular activities at the molecular level. These techniques facilitate the identification of early-stage cancers that might remain undetectable through traditional imaging methods. Innovative technologies such as reflectance confocal microscopy (RCM) and optical coherence tomography (OCT) which visualize skin at near-histologic detail and skin fluorescent imaging (SFI), which targets αvβ3 integrin expression, are promising for non-invasive early detection of melanoma. By integrating in vivo molecular imaging with tumor biomarkers, clinicians can gain more precise insights into processes integral to cancer biology, leading to improved diagnosis, prognosis and the development of personalized treatment strategies. This review explores imaging modalities used in skin cancer diagnosis, highlighting their advantages and limitations, with an emphasis on molecular imaging, stressing its potential to improve early detection, personalize treatment and monitor therapeutic responses.

Failure of machine learning models to generalize to new data is a core problem limiting the reliability of AI systems, partly due to the lack of simple and robust methods for comparing new data to the original training dataset. We propose a standardized approach for assessing data similarity in a model-agnostic manner by constructing a supervised autoencoder for generalizability estimation (SAGE). We compare points in a low-dimensional embedded latent space, defining empirical probability measures for k-Nearest Neighbors (kNN) distance, reconstruction of inputs and task-based performance. As proof of concept for classification tasks, we use MNIST and CIFAR-10 to demonstrate how an ensemble output probability score can separate deformed images from a mixture of typical test examples, and how this SAGE score is robust to transformations of increasing severity. As further proof of concept, we extend this approach to a regression task using non-imaging data (UCI Abalone). In all cases, we show that out-of-the-box model performance increases after SAGE score filtering, even when applied to data from the model's own training and test datasets. Our out-of-distribution scoring method can be introduced during several steps of model construction and assessment, leading to future improvements in responsible deep learning implementation.

PURPOSE: Despite the frequency with which patients with cancer receive radiotherapy, integrating radiation oncology data with other aspects of the clinical record remains challenging because of siloed and variable software systems, high data complexity, and inconsistent data encoding. Recognizing these challenges, the Veterans Affairs (VA) National Radiation Oncology Program (NROP) is developing Granular Radiotherapy Information Database (GRID), a platform and pipeline to combine radiotherapy data across the VA with the goal of both better understanding treatment patterns and outcomes and enhancing research and data analysis capabilities. METHODS: This study represents a proof-of-principle retrospective cohort analysis and review of select radiation treatment data from the VA Radiation Oncology Quality Surveillance Program (VAROQS) initiative. Key radiation oncology data elements were extracted from Digital Imaging and Communications in Medicine Radiotherapy extension (DICOM-RT) files and combined into a single database using custom scripts. These data were transferred to the VA's Corporate Data Warehouse (CDW) for integration and comparison with the VA Cancer Registry System and tumor sequencing data. RESULTS: The final cohort includes 1,568 patients, 766 of whom have corresponding DICOM-RT data. All cases were successfully linked to the CDW; 18.8% of VAROQS cases were not reported in the existing VA cancer registry. The VAROQS data contributed accurate radiation treatment details that were often erroneous or missing from the cancer registry record. Tumor sequencing data were available for approximately 5% of VAROQS cases. Finally, we describe a clinical dosimetric analysis leveraging GRID. CONCLUSION: NROP's GRID initiative aims to integrate VA radiotherapy data with other clinical data sets. It is anticipated to generate the single largest collection of radiation oncology–centric data merged with detailed clinical and genomic data, primed for large-scale quality assurance, research reuse, and discovery science.