MLMatthew D. Lee

Matthew D. Lee

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

Matthew D. Lee has a total of 680 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Enhancing 1p/19q Classification in Brain Gliomas Using Isocitrate Dehydrogenase Status: A Deep Learning Study

    by Jason E. Bowerman, Ashwath S. Kapilavai, Benjamin C. Wagner, Nghi C.D. Truong, James M. Holcomb, Divya D. Reddy, Niloufar Saadat, Kimmo J. Hatanpaa, Toral R. Patel, Baowei Fei, et al.

    IDH mutation & 1p/19q codeletion are critical biomarkers for glioma diagnosis & therapy. 1p/19q codeletion occurs exclusively in IDH-mutated gliomas. In this study, we developed a 2-stage, non-invasive, MRI-based deep learning method that leverages IDH status to enhance 1p/19q predictions. Multi-contrast brain tumor MRI & genomic information were obtained from five publicly available (TCIA, UCSF, EGD, UPenn & LGG), and three in-house/collaborator institutions (UTSW, NYU, UWM). Subjects were screened for the availability of IDH & 1p/19q status as well as T1, T1CE, T2, FLAIR MR images. For training purposes, missing T1 and FLAIR contrasts for the LGG database were generated using an in-house multi-contrast simulator. Two separate U-Nets were developed for 1p/19q-classification: a multi-contrast network (MC-Net) and a T2w-only network (T2-Net). A separate U-Net was developed for IDH classification (IDH-net). A total of 2044 subjects were used in training and testing IDH-Net, and 1426 subjects were used in training and testing the 1p/19q models. The IDH-Net was trained using subjects from TCIA, UTSW, and UPenn. The 1p/19q networks were trained using subjects from TCIA, UTSW, and LGG. The trained networks were tested on true held-out cases from NYU, UWM, EGD, and UCSF. In the 2-stage approach, subjects were initially classified for IDH status using IDH-Net. Predicted IDH-wildtype cases default to 1p/19q non-codeleted. Then the IDH-mutated cases were further classified for 1p/19q status using the 1p/19q-networks. IDH-Net achieved a classification accuracy of 93.7%. 1p/19q MC-Net & T2-Net achieved classification accuracies of 86.5% & 86.0%, respectively. In the 2-stage approach, 1p/19q MC-Net and T2-Net achieved accuracies of 91.5% & 91.2% respectively, improving the classification accuracy by ∼5%. This study demonstrates the effectiveness of leveraging IDH status to enhance 1p/19q classification. A ∼5% increase in classification accuracy was achieved when using the 2-stage approach, using IDH-Net to gate 1p/19q predictions. The developed method offers a reliable, non-invasive approach to determine important biomarkers for glioma diagnosis.

  • MRI and Clinical Features of Nonenhancing IDH -Wild-Type Glioblastomas: How to Make an Early Diagnosis and Distinguish from Mimics

    by James Ryan Loftus, Kanwar P. Singh, Sohil H. Patel, Matthew D. Lee, Matija Snuderl, Daniel Orringer, Rajan Jain

    A small subset of isocitrate dehydrogenase-wild-type (IDH-wt) glioblastomas (GBMs) initially present as nonenhancing, T2 FLAIR hyperintense cortical/superficial lesions on MRI, potentially leading to misdiagnosis on the initial imaging and hence delayed treatment. This study aimed to characterize the clinical and MRI features of nonenhancing IDH-wt GBMs to help radiologists in differentiating them from nonmalignant mimic diagnoses (eg, encephalitis). Additionally, the histologic, genomic, and survival profiles of nonenhancing GBMs were compared with those of enhancing GBMs. Clinical and MRI features from 32 patients, each with nonenhancing and enhancing GBMs, and 16 patients with nonmalignant mimic differential diagnoses from a single institution and publicly available data set were retrospectively analyzed. Imaging features were reviewed using the Visually Accessible Rembrandt Images features and the split ADC sign. χ2 tests and a binary logistic regression model were used to compare nonenhancing IDH-wt GBMs with nonmalignant mimics. Histopathologic and genomic analyses were performed on institutional cases. Overall survival between nonenhancing and enhancing GBMs was compared using Kaplan-Meier analysis. No significant difference in age, clinical presentation, or duration of symptoms was found between nonenhancing GBMs and nonmalignant mimics. Imaging features favoring nonenhancing GBMs included a greater proportion of non-contrast-enhancing tumor (OR, 7.4), larger anterior-posterior tumor dimension (OR, 8.4), restricted diffusion (OR, 3.6), and eloquent brain involvement (OR, 3.0) while features favoring mimics included greater edema (OR, 0.07), infiltrative T1 FLAIR ratio (OR, 0.68), hemorrhage (OR, 0.76), satellite lesions (OR, 0.84), and the split ADC sign (OR, 0.89). The logistic regression model achieved a mean area under the receiver operator characteristic curve of 0.89 (SD, 0.20) (accuracy 0.84, sensitivity 0.91, specificity 0.70, and precision 0.88). Twelve of 18 nonenhancing GBMs lacked histologic evidence of necrosis or microvascular proliferation ("molecular GBMs"). Genomic profiles were similar between nonenhancing and enhancing GBMs. Median overall survival was nonsignificantly longer in nonenhancing GBMs compared with enhancing GBMs (39 versus 21 months, P = .078). Nonenhancing GBMs demonstrate distinct MRI features that must be recognized for early diagnosis and differentiation from nonmalignant mimics. Nonenhancing GBMs demonstrated longer overall survival compared with enhancing GBMs, though they were not statistically significant.

  • Open Access

    Diagnosis of Multiple Sclerosis Using Multimodal Deep Learning Integrating Lesion and Normal-Appearing White Matter: A Retrospective Study with International Multicentre External Validation

    by Jiajian Ma, Valentin Stepanov, Wushuang Rui, Hsuan-Chih Chen, Maciej Lis, Aleksandra Stanek, Tomasz Puto, Michael Lan, Jenny Chen, Timothy Liu, et al.

    Current multiple sclerosis (MS) diagnosis relies primarily on focal white matter lesions (WMLs), which are frequently mimicked by other conditions. Normal-appearing white matter (NAWM) harbours complementary pathological information but remains clinically underutilised because NAWM alterations are macroscopically occult on routine scans and require non-routine quantitative imaging to visualise. Here, we show that NAWM-related diagnostic information can be recovered from routine structural MRI using a cross-modal deep-learning model. We developed DeepMS, a model co-trained on diffusion and structural MRI that operates solely on structural MRI at deployment. DeepMS achieved ROC-AUCs of 0.968 internally (n=837) and 0.940–0.974 across two international external cohorts (n=293 and n=1,756). In a multi-reader study, DeepMS outperformed the 2024 McDonald criteria imaging biomarkers. DeepMS retained robust performance after digital lesion removal and exhibited NAWM-dominant activation maps. Combined with established imaging biomarkers, DeepMS improved sensitivity (92.1% vs 74.8%) while maintaining high specificity (95.6% vs 92.3%) compared with corresponding biomarker composite based on the 2024 McDonald criteria. By decoding latent NAWM signals from routine scans and integrating them with WML features, this framework can potentially advance MS diagnosis beyond the current lesion-centric paradigm.

  • Open Access

    Correction to: MRI-based prediction of DNA methylation grade in IDH-mutant astrocytomas using qualitative imaging features and tumor volumetrics

    by Kanwar Partap Bir Singh, Matthew D. Lee, Matthew G. Young, Daniel Orringer, Yuxiu Wang, Matija Snuderl, Rajan Jain

  • MRI-based prediction of DNA methylation grade in IDH-mutant astrocytomas using qualitative imaging features and tumor volumetrics

    by Kanwar Partap Bir Singh, Matthew D. Lee, Matthew G. Young, Daniel Orringer, Yuxiu Wang, Matija Snuderl, Rajan Jain

    Purpose Histopathological grading of IDH-mutant astrocytomas demonstrates limited prognostic accuracy. However, DNA methylation subclassification has demonstrated improved prognostication beyond histological grading. This study aimed to investigate the associations between imaging features, tumor volumetric data, and DNA methylation grade in IDH-mutant astrocytomas. Methods We analyzed imaging features and volumetric data for 72 patients diagnosed with IDH-mutant astrocytomas, who underwent preoperative MRI and DNA methylation profiling. VASARI features and multicompartmental volumetrics were evaluated. Logistic regression was used to identify imaging predictors of methylation subclass, WHO histologic grade, copy number variation (CNV), and CDKN2A/B homozygous deletion. Univariable and multivariable Cox proportional hazard models were also developed to assess these variables’ influence on overall survival and progression-free survival. Results Patients were classified into 27 methylation high-grade (A_IDH_HG) and 45 methylation low-grade (A_IDH_LG) tumors. Tumor volumes and proportions varied by methylation grade, CNV status, and WHO histologic grade, but not by CDKN2A/B status. Imaging features distinguished methylation subclasses with 75% accuracy (AUC = 0.77). Methylation high-grade subclass was associated with imaging features such as midline crossing, ependymal extension, and poorly defined enhancing margins. Predictive performance for WHO histologic grade, CNV status, and CDKN2A/B deletion was moderate (AUC = 0.67, 0.69, and 0.65, respectively). Methylation grade, CDKN2A/B status, VASARI features, and proportions of edema and non-contrast enhancing tumor were significantly associated with survival. Conclusion MRI-derived imaging features facilitate noninvasive prediction of DNA methylation subclass in IDH-mutant astrocytomas.

  • Open Access

    Categorical and phenotypic image synthetic learning as an alternative to federated learning

    by Nghi C. D. Truong, Chandan Ganesh Bangalore Yogananda, Benjamin C. Wagner, James M. Holcomb, Divya D. Reddy, Niloufar Saadat, Jason Bowerman, Kimmo J. Hatanpaa, Toral R. Patel, Baowei Fei, et al.

    Multi-center collaborations are crucial in developing robust and generalizable machine learning models in medical imaging. Traditional methods, such as centralized data sharing or federated learning (FL), face challenges, including privacy issues, communication burdens, and synchronization complexities. We present CATegorical and PHenotypic Image SyntHetic learnING (CATphishing), an alternative to FL using Latent Diffusion Models (LDM) to generate synthetic multi-contrast three-dimensional magnetic resonance imaging data for downstream tasks, eliminating the need for raw data sharing or iterative inter-site communication. Each institution trains an LDM to capture site-specific data distributions, producing synthetic samples aggregated at a central server. We evaluate CATphishing using data from 2491 patients across seven institutions for isocitrate dehydrogenase mutation classification and three-class tumor-type classification. CATphishing achieves accuracy comparable to centralized training and FL, with synthetic data exhibiting high fidelity. This method addresses privacy, scalability, and communication challenges, offering a promising alternative for collaborative artificial intelligence development in medical imaging.


  • Top Cited Papers
    Open Access

    Towards fair decentralized benchmarking of healthcare AI algorithms with the Federated Tumor Segmentation (FeTS) challenge

    by Maximilian Zenk, Ujjwal Baid, Sarthak Pati, Akis Linardos, Brandon Edwards, Micah Sheller, Patrick Foley, Alejandro Aristizabal, David Zimmerer, Alexey Gruzdev, et al.

    Computational competitions are the standard for benchmarking medical image analysis algorithms, but they typically use small curated test datasets acquired at a few centers, leaving a gap to the reality of diverse multicentric patient data. To this end, the Federated Tumor Segmentation (FeTS) Challenge represents the paradigm for real-world algorithmic performance evaluation. The FeTS challenge is a competition to benchmark (i) federated learning aggregation algorithms and (ii) state-of-the-art segmentation algorithms, across multiple international sites. Weight aggregation and client selection techniques were compared using a multicentric brain tumor dataset in realistic federated learning simulations, yielding benefits for adaptive weight aggregation, and efficiency gains through client sampling. Quantitative performance evaluation of state-of-the-art segmentation algorithms on data distributed internationally across 32 institutions yielded good generalization on average, albeit the worst-case performance revealed data-specific modes of failure. Similar multi-site setups can help validate the real-world utility of healthcare AI algorithms in the future.


  • Mitigating data scarcity in the classification of glioma molecular subtypes: the power of generative imaging (Erratum)

    by Nghi C. D. Truong, Chandan Ganesh Bangalore Yogananda, Benjamin C. Wagner, Niloufar Saadat, James M. Holcomb, Divya Reddy, Sadeem Lodhi, Jason Bowerman, Kimmo J. Hatanpaa, Toral R. Patel, et al.

    Publisher's Note: This paper, originally published on 10 April 2025, was replaced with a corrected/revised version on 23 June 2025. If you downloaded the original PDF but are unable to access the revision, please contact SPIE Digital Library Customer Service for assistance.

  • Mitigating data scarcity in the classification of glioma molecular subtypes: the power of generative imaging

    by Nghi C D Truong, Chandan Ganesh Bangalore Yogananda, Benjamin C Wagner, Niloufar Saadat, James M Holcomb, Divya Reddy, Sadeem Lodhi, Jason Bowerman, Kimmo J Hatanpaa, Toral R Patel, et al.

    Isocitrate dehydrogenase (IDH) mutation status is a critical prognostic indicator in glioma patients. Numerous studies have focused on developing non-invasive methodologies to classify IDH status using pre-operative MRI scans. However, the challenge lies in data scarcity and class imbalance in IDH mutations. This study explores generative AI methods to augment training data and enhance IDH classification accuracy. We developed a 3D conditional latent diffusion model (LDM) for generating 3D multi-contrast brain tumor MRI data (128×128×64 with a voxel spacing of 1.5×1.5×2.0mm) with whole tumor mask and IDH mutation status as conditions. The LDM comprises a 3D autoencoder for perceptual compression and a conditional 3D diffusion model (DM) for generating multi-contrast synthetic samples guided by tumor masks and the IDH mutation status. We incorporated two types of attention modules within the denoising UNet of the LDM to capture the semantic class-dependent data distribution driven by the provided whole tumor mask and IDH status. The LDM was trained using two brain tumor datasets: The Cancer Genome Atlas dataset and an internal dataset from the University of Texas Southwestern Medical Center. The synthetic images generated by the LDM were then used to train IDH classification models, which were subsequently tested on real brain tumor data comprising 327 mutated and 1,394 wild-type cases from the University of California San Francisco Preoperative Diffuse Glioma MRI dataset, the Erasmus Glioma Database, the University of Pennsylvania glioblastoma, and two held-out internal datasets. The IDH classification models, trained on synthetic images and tested on real data, achieved an excellent overall classification accuracy of 94.02%. This approach has the potential to be extended to other molecular markers where data scarcity presents a challenge.


  • Machine learning-based prognostic subgrouping of glioblastoma: A multicenter study

    by Hamed Akbari, Spyridon Bakas, Chiharu Sako, Anahita Fathi Kazerooni, Javier Villanueva-Meyer, Jose A Garcia, Elizabeth Mamourian, Fang Liu, Quy Cao, Russell T Shinohara, et al.

    Background: Glioblastoma (GBM) is the most aggressive adult primary brain cancer, characterized by significant heterogeneity, posing challenges for patient management, treatment planning, and clinical trial stratification. Methods: We developed a highly reproducible, personalized prognostication, and clinical subgrouping system using machine learning (ML) on routine clinical data, magnetic resonance imaging (MRI), and molecular measures from 2838 demographically diverse patients across 22 institutions and 3 continents. Patients were stratified into favorable, intermediate, and poor prognostic subgroups (I, II, and III) using Kaplan–Meier analysis (Cox proportional model and hazard ratios [HR]). Results: The ML model stratified patients into distinct prognostic subgroups with HRs between subgroups I–II and I–III of 1.62 (95% CI: 1.43–1.84, P < .001) and 3.48 (95% CI: 2.94–4.11, P < .001), respectively. Analysis of imaging features revealed several tumor properties contributing unique prognostic value, supporting the feasibility of a generalizable prognostic classification system in a diverse cohort. Conclusions: Our ML model demonstrates extensive reproducibility and online accessibility, utilizing routine imaging data rather than complex imaging protocols. This platform offers a unique approach to personalized patient management and clinical trial stratification in GBM.