Publications

This page highlights a selection of publications from the lab. For a complete list of publications, please see the links below.

Selected Publications

2025

Zhuang L, Tabatabaei SMH, Salehi-Rad R, Tran LM, Aberle DR, Prosper AE, Hsu W. Vision-language model-based semantic-guided imaging biomarker for lung nodule malignancy prediction. J Biomed Inform. 2025 Oct 27:104947.

💡 This work emphasizes the importance of domain knowledge in AI, enabling the extraction of clinically meaningful deep imaging biomarkers and exploring vision-language models as a means of integrating semantic features for malignancy prediction of pulmonary nodules.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/41161557/
💻 Code: https://github.com/luotingzhuang/CLIP_nodule

Lotter W, Hippe DS, Oshiro T, Lowry KP, Milch HS, Miglioretti DL, Elmore JG, Lee CI, Hsu W. Influence of Mammography Acquisition Parameters on AI and Radiologist Interpretive Performance. Radiol Artif Intell. 2025 Sep 17:e240861.

💡 In this study, we examined how technical acquisition parameters in mammography, like x-ray exposure and compression force, impact diagnostic performance differently for AI models and radiologists in an analysis that included over 28,000 screening mammograms.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/40960399/

Zhuang L, Park SH, Skates SJ, Prosper AE, Aberle DR, Hsu W. Advancing Precision Oncology Through Modeling of Longitudinal and Multimodal Data. IEEE Rev Biomed Eng. 2025 Jul 3;PP.

💡 A review of methods for integrating longitudinal and multimodal data to better capture the complexity of cancer progression. The review highlights challenges in data acquisition, integration, model interpretability, and evaluation, emphasizing the need for collaborative efforts and innovative methodologies to fully realize the potential of longitudinal multimodal analysis in cancer care.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/40608875/
📎 Supplement: https://doi.org/10.1109/RBME.2025.3577587/mm1

Ding R, Luong KD, Rodriguez E, da Silva ACAL, Hsu W. Combining graph neural network and Mamba to capture local and global tissue spatial relationships in whole slide images. Sci Rep. 2025 May 25;15(1):18261.

💡 We proposed GAT-Mamba, a novel deep learning model combining GNNs and Mamba state space models to predict progression-free survival in early-stage lung adenocarcinoma using whole-slide pathology images. GAT-Mamba captures both local and global spatial relationships among tissue tiles, outperforming existing models in accuracy and efficiency.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/40415116/
💻 Code: https://github.com/rina-ding/gat-mamba

Zhang T, Ding R, Luong KD, Hsu W. Evaluating an information theoretic approach for selecting multimodal data fusion methods. J Biomed Inform. 2025 Jul;167:104833.

💡 We conducted an analysis on partial information decomposition (PID) metrics in guiding multimodal data fusion for biomedical applications. While PID metrics, which quantify redundancy, uniqueness, and synergy between data modalities, can provide useful insights for modeling, we found that these metrics alone do not consistently predict optimal model performance across diverse datasets.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/40354908/
💻 Code: https://github.com/zhtyolivia/pid-multimodal

Yu TT, Hoyt AC, Joines MM, Fischer CP, Yaghmai N, Chalfant JS, Chow L, Mortazavi S, Sears CD, Sayre J, Elmore JG, Hsu W, Milch HS. Mammographic classification of interval breast cancers and artificial intelligence performance. J Natl Cancer Inst. 2025 Aug 1;117(8):1627-1638.

💡 We looked at interval breast cancers that appear between regular mammogram screenings and whether AI could help catch them earlier. The AI tool was able to flag many of these cases, especially when the cancer was faintly visible, suggesting it could help radiologists spot warning signs they might otherwise overlook.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/40249223/

2024

Lin Y, Hoyt AC, Manuel VG, Inkelas M, Hsu W. Using Discrete Event Simulation to Design and Assess an AI-aided Workflow for Same-day Diagnostic Testing of Women Undergoing Breast Screening. AMIA Jt Summits Transl Sci Proc. 2024 May 31;2024:314-323.

💡 AI-aided same-day diagnostic workups for abnormal mammograms can reduce patient anxiety and unnecessary recalls but may also decrease daily patient volume by 4% and increase clinic time by 24%, simulations show. Adjusting hours and resources could help. Discrete event simulations can help with planning for these unintended consequences.
📄 Paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11141813/

Ramwala OA, Lowry KP, Cross NM, Hsu W, Austin CC, Mooney SD, Lee CI. Establishing a Validation Infrastructure for Imaging-Based Artificial Intelligence Algorithms Before Clinical Implementation. J Am Coll Radiol. 2024 May 22:S1546-1440(24)00451-4.

💡 AI algorithms, even FDA-cleared ones, need local validation to ensure accuracy and safety. Challenges include data privacy and ensuring models perform consistently across diverse populations. In this manuscript, we propose steps for establishing an efficient validation infrastructure to enhance radiology workflows and patient outcomes.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/38789066/

Rahrooh A, Garlid AO, Bartlett K, Coons W, Petousis P, Hsu W, Bui AAT. Towards a framework for interoperability and reproducibility of predictive models. J Biomed Inform. 2024 Jan;149:104551.

💡 Machine learning models in healthcare are not disseminated in a standardized manner, hindering reproducibility. We describe an automated metadata pipeline that converts models into extended Predictive Modeling Markup Language files, thereby enhancing interoperability and reproducibility.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/38000765/

2023

Wei L, Yadav A, Hsu W. CTFlow: Mitigating Effects of Computed Tomography Acquisition and Reconstruction with Normalizing Flows. Med Image Comput Comput Assist Interv. 2023 Oct;14226:413-422.

💡 This paper introduces CTFlow, a method using normalizing flows to harmonize CT scans acquired under different imaging conditions, varying radiation doses and reconstruction kernels. Unlike traditional approaches that produce a single corrected image, CTFlow models the full distribution of plausible reconstructions, providing more flexibility and awareness of uncertainty. It outperformed GAN-based methods in both image quality metrics and consistency of lung nodule detection across 186 real-world low-dose CT scans.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/38737498/
💻 Code: https://github.com/hsu-lab/ctflow

Prosper AE, Kammer MN, Maldonado F, Aberle DR, Hsu W. Expanding Role of Advanced Image Analysis in CT-detected Indeterminate Pulmonary Nodules and Early Lung Cancer Characterization. Radiology. 2023;309(1):e222904.

💡 Implementing low-dose chest CT for lung screening can advance lung cancer care through early detection. However, millions of pulmonary nodules will be detected annually, which must be managed. Accurate image analysis using AI/ML is key to minimizing potential harms that may result from low-risk nodules. Challenges to the clinical utility of AI/ML include the heterogeneity of imaging parameters and the need for well-annotated datasets.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/37815447/

Lin Y, Liang LJ, Ding R, Prosper AE, Aberle DR, Hsu W. Factors Associated With Nonadherence to Lung Cancer Screening Across Multiple Screening Time Points. JAMA Netw Open. 2023 May 1;6(5):e2315250.

💡 Low adherence to Lung-RADS recommendations in practice contrasts with high adherence in trials. Identifying nonadherent patients is crucial for realizing the mortality reduction observed in randomized trials. Our study identified temporal factors associated with nonadherence, enabling us to develop targeted interventions to enhance adherence.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/37227725/

Marasinou C, Li B, Paige J, Omigbodun A, Nakhaei N, Hoyt A, Hsu W. Improving the Quantitative Analysis of Breast Microcalcifications: A Multiscale Approach. J Digit Imaging. 2023 Jun;36(3):1016-1028.

💡This study presents a multiscale, two-stage approach for segmenting breast microcalcifications (MCs) in 2D digital mammograms, which combines Hessian-based blob detection with a regression convolutional neural network trained to localize MCs using proximity functions. In a downstream classification task, features extracted from the segmented MCs improved the ability to distinguish benign from malignant amorphous calcifications, outperforming baseline methods with an AUC of 0.763.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/36820930/
💻 Code: https://github.com/cmarasinou/HDoGReg

Ding R, Yadav A, Rodriguez E, Silva ACAL da, Hsu W. Tailoring pretext tasks to improve self-supervised learning in histopathologic subtype classification of lung adenocarcinomas. Comput Biol Med. 2023;166:107484.

💡This paper proposes three tailored self-supervised learning (SSL) tasks to improve histologic subtype classification of lung adenocarcinoma using whole-slide pathology images. Unlike generic SSL methods, these tasks are designed to mimic how pathologists interpret tissue, by learning spatial relationships across magnifications and predicting eosin staining from hematoxylin images to capture cytoplasmic features. The results suggest that designing SSL tasks aligned with domain-specific diagnostic goals can enhance model accuracy while reducing reliance on expert labeling.

📄 Paper: https://pubmed.ncbi.nlm.nih.gov/37741228/
💻 Code: https://github.com/rina-ding/ssl_luad_classification

Paige JS, Lee CI, Wang PC, Hsu W, Brentnall AR, Hoyt AC, Naeim A, Elmore JG. Variability Among Breast Cancer Risk Classification Models When Applied at the Level of the Individual Woman. J Gen Intern Med. 2023 Aug;38(11):2584-2592.

💡 This study evaluated the performance of three widely used breast cancer risk models (BCRAT (Gail), BCSC, and IBIS) when applied to individual women rather than populations. Although all three models showed similar accuracy at the population level, they often disagreed on whether a specific woman was “high risk,” especially when using the common 1.67% 5-year risk threshold. The findings highlight the limitations of current risk models in personalized care and underscore the need for more precise tools to guide individual decision-making.
📄 Paper: https://pubmed.ncbi.nlm.nih.gov/36749434/

2022

Marathe K, Marasinou C, Li B, Nakhaei N, Li B, Elmore JG, et al. Automated quantitative assessment of amorphous calcifications: Towards improved malignancy risk stratification. Comput Biol Med. 2022;146:105504. Available from: https://doi.org/10.1016%2Fj.compbiomed.2022.105504

Hsu W, Hippe DS, Nakhaei N, Wang PC, Zhu B, Siu N, et al. External Validation of an Ensemble Model for Automated Mammography Interpretation by Artificial Intelligence. JAMA Network Open. 2022;5(11):e2242343. Available from: https://doi.org/10.1001%2Fjamanetworkopen.2022.42343

Lin Y, Fu M, Ding R, Inoue K, Jeon C, Hsu W, et al. Patient Adherence to Lung CT Screening Reporting & Data System–Recommended Screening Intervals in the United States: A Systematic Review and Meta-Analysis. J Thorac Oncol. 2022;17(1):38–55.

Hendrix N, Lowry KP, Elmore JG, Lotter W, Sorensen G, Hsu W, et al. Radiologist Preferences for Artificial Intelligence-Based Decision Support During Screening Mammography Interpretation. Journal of the American College of Radiology. 2022;19(10):1098–110. Available from: http://dx.doi.org/10.1016/j.jacr.2022.06.019

Hsu W, Sohn JH. Using Radiomics for Risk Stratification: Where We Need to Go. Radiology. 2022;302(2):435–7. Available from: https://doi.org/10.1148%2Fradiol.2021212085

Older

Prosper A, Inoue K, Brown K, Bui A, Aberle D, Hsu W. Association of Inclusion of More Black Individuals in Lung Cancer Screening With Reduced Mortality. Jama Netw Open. 2021;4(8):e2119629.

Peterson E, May F, Kachikian O, Soroudi C, Naini B, Kang Y, et al. Automated identification and assignment of colonoscopy surveillance recommendations for individuals with colorectal polyps. Gastrointest Endosc. 2021;94(5):978–87.

Matiasz N, Wood J, Wang W, Silva A, Hsu W. Experiment Selection in Meta-Analytic Piecemeal Causal Discovery. Ieee Access. 2021;9:97929–41.

Emaminejad N, Wahi‐Anwar M, Kim G, Hsu W, Brown M, McNitt‐Gray M. Reproducibility of lung nodule radiomic features: Multivariable and univariable investigations that account for interactions between CT acquisition and reconstruction parameters. Med Phys. 2021;48(6):2906–19.

Smedley N, Aberle D, Hsu W. Using deep neural networks and interpretability methods to identify gene expression patterns that predict radiomic features and histology in non-small cell lung cancer. J Medical Imaging. 2021;8(3):031906.

Smedley NF, El-Saden S, Hsu W. Discovering and interpreting transcriptomic drivers of imaging traits using neural networks. Bioinform. 2020;36(11):3537–48. Available from: http://dx.doi.org/10.1093/bioinformatics/btaa126

Li M, Hsu W, Xie X, Cong J, Gao W. SACNN: Self-Attention Convolutional Neural Network for Low-Dose CT Denoising With Self-Supervised Perceptual Loss Network. Ieee T Med Imaging. 2020;39(7):2289–301. Available from: http://dx.doi.org/10.1109/tmi.2020.2968472

Gao Y, Kalbasi A, Hsu W, Ruan D, Fu J, Shao J, et al. Treatment effect prediction for sarcoma patients treated with preoperative radiotherapy using radiomics features from longitudinal diffusion-weighted MRI. Phys Medicine Biology. 2020;65(17):175006. Available from: https://doi.org/10.1088%2F1361-6560%2Fab9e58

Shen S, Han SX, Aberle DR, Bui AA, Hsu W. An interpretable deep hierarchical semantic convolutional neural network for lung nodule malignancy classification. Expert Systems with Applications. 2019;128:84–95. Available from: https://doi.org/10.1016%2Fj.eswa.2019.01.048

Winter A, Aberle DR, Hsu W. External validation and recalibration of the Brock model to predict probability of cancer in pulmonary nodules using NLST data. Thorax. 2019;74(6):551–63. Available from: https://doi.org/10.1136%2Fthoraxjnl-2018-212413

Hsu W, Elmore JG. Shining Light Into the Black Box of Machine Learning. JNCI: Journal of the National Cancer Institute. 2019;111(9):877–9. Available from: http://dx.doi.org/10.1093/jnci/djy226

Petousis P, Winter A, Speier W, Aberle D, Hsu W, Bui A. Using Sequential Decision Making to Improve Lung Cancer Screening Performance. Using Sequential Decision Making to Improve Lung Cancer Screening Performance. 2019;7(119403–119419).

Hsu W, Hoyt AC. Using Time as a Measure of Impact for AI Systems: Implications in Breast Screening. Radiology Artif Intell. 2019;1(4):e190107. Available from: http://dx.doi.org/10.1148/ryai.2019190107

Page updated

Google Sites

Report abuse