About

Patrick Bolan, PhD, is an Associate Professor in the Department of Radiology at the University of Minnesota. He serves as Associate Medical Director at the Center for Clinical Imaging Research and as Associate Director for the Center for Radiology Research Resources. Dr. Bolan's academic journey began with a B.S. in Mechanical Engineering from the University of Illinois and postgraduate studies at UC Berkeley. After working in industry as a software engineer for five years, he joined the University of Minnesota in 1999, where he developed methods for quantitative MR spectroscopy of breast cancer and earned his Ph.D. in Biomedical Engineering in 2003. His research focuses on developing quantitative magnetic resonance imaging (MRI) methods and integrating advanced imaging techniques into clinical trials related to cancer and obesity. He has continued his work at the UMN Center for Magnetic Resonance Research as a postdoctoral researcher, assistant, and associate professor, contributing significantly to the field of medical physics and MRI technology.

Research topics

• Medicine
• Artificial Intelligence
• Radiology
• Oncology
• Nuclear medicine
• Internal medicine

Selected publications

• Agentic LLM Workflow for MR Spectroscopy Volume-of-Interest Placements in Brain Tumors

  ArXiv.org · 2026-03-09

  articleOpen accessSenior author

  Magnetic resonance spectroscopy (MRS) provides clinically valuable metabolic characterization of brain tumors, but its utility depends on accurate placement of the spectroscopy volume-of-interest (VOI). However, VOI placement typically has a broad operating window: for a given tumor there are multiple possible VOIs that would lead to high-quality MRS measurements. Thus, a VOI place-ment can be tuned for clinician preference, case-specific anatomy, and clinical pri-orities, which leads to high inter-operator variability, especially for heterogeneous tumors. We propose an agentic large language model (LLM) workflow that de-composes VOI placement into generation of diverse candidate VOIs, from which the LLM selects an optimal one based on quantitative metrics. Candidate VOIs are generated by vision transformer-based placement models trained with differ-ent objective function preferences, which allows selection from acceptable alterna-tives rather than a single deterministic placement. On 110 clinical brain tumor cas-es, the agentic workflow achieves improved solid tumor coverage and necrosis avoidance depending on the user preferences compared to the general-purpose expert placements. Overall, the proposed workflow provides a strategy to adapt VOI placement to different clinical objectives without retraining task-specific models.

  PublisherOA PDF

• Agentic LLM Workflow for MR Spectroscopy Volume-of-Interest Placements in Brain Tumors

  arXiv (Cornell University) · 2026-03-09

  preprintOpen accessSenior author

  Magnetic resonance spectroscopy (MRS) provides clinically valuable metabolic characterization of brain tumors, but its utility depends on accurate placement of the spectroscopy volume-of-interest (VOI). However, VOI placement typically has a broad operating window: for a given tumor there are multiple possible VOIs that would lead to high-quality MRS measurements. Thus, a VOI place-ment can be tuned for clinician preference, case-specific anatomy, and clinical pri-orities, which leads to high inter-operator variability, especially for heterogeneous tumors. We propose an agentic large language model (LLM) workflow that de-composes VOI placement into generation of diverse candidate VOIs, from which the LLM selects an optimal one based on quantitative metrics. Candidate VOIs are generated by vision transformer-based placement models trained with differ-ent objective function preferences, which allows selection from acceptable alterna-tives rather than a single deterministic placement. On 110 clinical brain tumor cas-es, the agentic workflow achieves improved solid tumor coverage and necrosis avoidance depending on the user preferences compared to the general-purpose expert placements. Overall, the proposed workflow provides a strategy to adapt VOI placement to different clinical objectives without retraining task-specific models.

  PublisherDOI

• 3D GRASE with Long Echo Trains and Inner Volume Excitation

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025-09-16

  article

  Motivation: High-resolution 3D first spin echo (FSE) with long echo trains may require long scan times because of 3D k-space coverage. Goal(s): Our goal was to accelerate the high-resolution 3D FSE acquisition. Approach: We implemented a GRASE-type (gradient- and spin-echo) acquisition and inner volume excitation to reduce scan times. Results: The GRASE-type data acquisition enabled more rapid k-space sampling, while the impact of the increased sensitivity to off-resonance/magnetic susceptibility effects was minor at 3T. Inner volume excitation was useful to reduce the scan time while maintaining spatial resolution or to increase the spatial resolutions while maintaining scan time. Impact: 3D high-resolution FSE acquisitions have been accelerated by combining GRASE acquisitions, long echo trains and inner volume excitation, allowing efficient 3D k-space sampling for rapid high-resolution imaging.

  PublisherDOI

• Advanced DWI Models for Early Prediction of Pathologic Complete Response in Breast Cancer Treatment: Preliminary Results in a Multicenter Trial

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025-09-16

  article

  Motivation: DWI shows promise for early prediction of pathologic complete response (pCR), an important prognostic marker in neoadjuvant chemotherapy (NAC) treatment. Goal(s): To compare the performance of conventional ADC and advanced DWI models in predicting pCR in an ongoing multicenter prospective trial. Approach: In 25 women undergoing NAC, MRI scans were performed at pre- and early-treatment (post 1-cycle) timepoints. Tumors were evaluated using IVIM, DKI, and RSI models, and compared with ADC. Results: Changes in ADC, IVIM, DKI, and RSI metrics predicted pCR at early-treatment with moderately-high AUCs (0.77-0.92), suggesting future potential utility in guiding breast cancer treatment. Impact: ADC and advanced DWI models demonstrate accurate early prediction of pathologic response in breast tumors undergoing neoadjuvant chemotherapy. Results support their potential as imaging biomarkers to help optimize breast cancer treatments in the future.

  PublisherDOI

• Cross-vendor repeatability and reproducibility of quantitative MRI for prostate cancer applications

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025-09-16

  article

  Motivation: Variability in quantitative MRI (qMRI) parameter estimates across different vendor's systems can hinder translation of predictive modeling approaches. Goal(s): To develop cross-vendor protocols and perform quantitative phantom and in vivo measurements to assess repeatability and reproducibility. Approach: MRI data was acquired on phantoms and two volunteers on three sites each with a different vendor's systems to assess multiple qMRI parameters including T1, T2 and ADC. Results: Cross-vendor protocols were implemented, and initial phantom and in vivo studies have been performed. The coefficient of variation within and between sites was assessed. Impact: This work demonstrates initial results at characterizing repeatability and reproducibility of quantitative MRI (qMRI) across vendors and sites, a critical step needed to harmonize results prior to evaluating their use in developed computer-aided diagnostic tools for prostate cancer.

  PublisherDOI

• Brief Report: Safety of Pulse-Dose Osimertinib for Treatment of Leptomeningeal Disease or Refractory Brain Metastases in EGFR-Mutated Nonsmall Cell Lung Cancer

  Clinical Lung Cancer · 2025-08-06

  articleOpen access
  PublisherOA PDFDOI

• Visceral adipose tissue in obesity: A comparison between DXA and MRI measures matched for anatomical location

  Physiology · 2025-05-01

  article

  Introduction: Visceral adipose tissue (VAT) accumulation is associated with an increased risk of cardiovascular and metabolic disease in individuals. VAT can be accurately measured by magnetic resonance imaging (MRI); however, this technology is costly and time consuming. Dual X-ray absorptiometry (DXA) presents an alternate method to estimate VAT. Traditionally, VAT measurements using DXA and MRI are taken at different anatomical locations. Although these methods correlate, their comparative accuracy remains unconfirmed. This study aimed to compare VAT volume measurements between DXA and MRI at matched anatomical locations, hypothesizing that the two methods would yield similar results Methods: A total of fifty-one MRI and DXA scans were analyzed from participants with obesity and without diabetes (18-65 years, BMI range of ≥30 to ≤55 kg/m 2 ). VAT was measured at the L3 vertebral body region by manual manipulation of the android region of interest (ROI) within the DXA scan (Hologic Inc., Marlborough, MA, USA), and manual segmentation of the MRI-VAT (Siemens Prisma, Erlangen, Germany). MRI-VAT was expressed as total segmented volume and as fractional volume (corrected for the fat fraction in each voxel). Paired t-tests assessed differences between the DXA-VAT and MRI-VAT volumes. Pearson’s correlation coefficients were computed comparing VAT measures from DXA and MRI. Bland-Altman tests assessed magnitude of bias and 95% limits of agreement between measures. Data were analyzed using RStudio version 4.4.1, and significance was set at p<0.05. Results: There were no significant differences between DXA-VAT volume and MRI-VAT total volume (780.9±251.5mL vs. 727.9±318.0mL, p=0.088). However, there was significant (p=0.014) difference between DXA-VAT volume and MRI-VAT fractional volume (780.9±251.5mLvs. 640.8±311.5L). DXA-VAT volume and MRI-VAT total volume showed a positive and strong correlation (R=0.731, p<0.001), and DXA-VAT volume and MRI-VAT fractional volume showed similar values (R=0.737, p<0.001). The Bland-Atman analysis showed a positive bias (+53.04cm 3 ) of DXA-VAT volume relative to MRI-VAT total volume and good overall limits of agreement (LoA= 479.93, -373.84). Similar agreement was observed between a positive bias (+140.1cm3) of DXA-VAT volume and MRI-VAT fractional volume with good LoA (554.6, -247.3). Conclusion: When comparing VAT measurements at the L3 vertebral level, we found strong agreement between DXA and MRI VAT volume measurements, though DXA VAT were consistently higher than MRI VAT measurements. Study was supported by UMN CTSA Award (UM1TR004405-01A1) Center for Advancing Translational Sciences and NIH (R01DK124484 to LSC, P41EB027061, S10OD017974 for MRI support). This abstract was presented at the American Physiology Summit 2025 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.

  PublisherDOI

• Ketogenesis mitigates metabolic dysfunction–associated steatotic liver disease through mechanisms that extend beyond fat oxidation

  Journal of Clinical Investigation · 2025-04-24 · 12 citations

  articleOpen access

  The progression of metabolic dysfunction-associated steatotic liver disease (MASLD) to metabolic dysfunction-associated steatohepatitis (MASH) involves alterations in both liver-autonomous and systemic metabolism that influence the liver's balance of fat accretion and disposal. Here, we quantify the contributions of hepatic oxidative pathways to liver injury in MASLD-MASH. Using NMR spectroscopy, UHPLC-MS, and GC-MS, we performed stable isotope tracing and formal flux modeling to quantify hepatic oxidative fluxes in humans across the spectrum of MASLD-MASH, and in mouse models of impaired ketogenesis. In humans with MASH, liver injury correlated positively with ketogenesis and total fat oxidation, but not with turnover of the tricarboxylic acid cycle. Loss-of-function mouse models demonstrated that disruption of mitochondrial HMG-CoA synthase (HMGCS2), the rate-limiting step of ketogenesis, impairs overall hepatic fat oxidation and induces an MASLD-MASH-like phenotype. Disruption of mitochondrial β-hydroxybutyrate dehydrogenase (BDH1), the terminal step of ketogenesis, also impaired fat oxidation, but surprisingly did not exacerbate steatotic liver injury. Taken together, these findings suggest that quantifiable variations in overall hepatic fat oxidation may not be a primary determinant of MASLD-to-MASH progression, but rather that maintenance of ketogenesis could serve a protective role through additional mechanisms that extend beyond overall rates of fat oxidation.

  PublisherOA PDFDOI

• Restricted Diffusion Volume (RDV): a novel quantitative metric to improve prediction of breast cancer response to neoadjuvant chemotherapy

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025-09-16

  article1st authorCorresponding

  Motivation: Diffusion-weighted imaging can be used to monitor treatment response in breast cancer. The conventional approach of measuring ADC requires manual processing and moderate predictive performance early in treatment. Goal(s): To develop an objective measurement of diffusion response and evaluate its predictive performance. Approach: Retrospective analyses were performed using data from the ACRIN 6698 trial. The restricted diffusion volume (RDV) was calculated by summing pixels meeting multiple objective criteria, including ADC in the range [0.6, 1.4] x10-3 mm2/s. The predictive performance of RDV was compared to conventional ADC and contrast-enhanced imaging. Results: RDV outperformed ADC early in treatment and matched its performance post-treatment. Impact: RDV is a novel, semi-automated diffusion metric that provides improved performance over conventional ROI-mean ADC. This can be used to optimize treatment and could lead to contrast-free therapy monitoring.

  PublisherDOI

• Can we cut the contrast? DWI shows similar value to standard MRI for assessing treatment response in breast cancer

  European Radiology · 2025-08-07 · 1 citations

  letter1st authorCorresponding
  PublisherDOI

Recent grants

• NIH Grant R01CA120509

  NIH · $767k · 2010

• NIH Grant R21CA201834

  NIH · $326k · 2018

Frequent coauthors

• Savannah C. Partridge

  Memorial Sloan Kettering Cancer Center

  122 shared

• Helga S. Marques

  Cancer Research Center

  118 shared

• Nola M. Hylton

  University of California, San Francisco

  116 shared

• Thomas L. Chenevert

  University of Michigan–Ann Arbor

  114 shared

• Wen Li

  University of California, San Francisco

  107 shared

• David C. Newitt

  University Hospital Heidelberg

  104 shared

• Despina Kontos

  Columbia University

  104 shared

• Jon A. Steingrimsson

  104 shared

Labs

• Center for Magnetic Resonance ResearchPI