About

Muhammad Murtaza Hassan is an Assistant Professor of Chemistry with a background that includes an HBSc in Chemistry Specialist & Math Minor from the University of Toronto Mississauga, obtained in 2013, an MSc in Organic Chemistry from York University in 2017, and a PhD in Medicinal Chemistry from the University of Toronto Mississauga in 2021. He completed postdoctoral studies at Stanford University from 2021 to 2025. His research focuses on developing therapeutics at the interface of chemically induced proximity (CIP), covalent pharmacology, and structure-based rational drug design. His lab aims to create generalizable, rationally designed molecular glue drugs that can transform targeted cancer therapy by hijacking endogenous cellular machinery to induce targeted protein degradation, phosphorylation, acetylation, and more. The lab emphasizes the integration of covalent chemistry into structure-guided CIP strategies to expand the druggable proteome and develop highly potent, selective drugs capable of targeting disease-driving proteins such as oncoproteins in cancer. Hassan's work addresses the challenge of designing monovalent degraders or 'molecular glues' with improved drug-like properties, moving beyond traditional bivalent design approaches. His research is highly translational, with direct relevance to oncology and potential for commercialization, supported by collaborations with leading institutions and a commitment to fostering a diverse, collaborative, and innovative research environment.

Research topics

• Artificial Intelligence
• Computer Science
• Medicine
• Machine Learning
• Optics
• Statistics
• Surgery
• Physics
• Geometry
• Mathematics
• Pathology
• Psychology
• Radiology
• Ophthalmology

Selected publications

• A Data-Centric Framework for Intraoperative Fluorescence Lifetime Imaging for Glioma Surgical Guidance

  arXiv (Cornell University) · 2026-04-28

  preprintOpen access

  Accurate intraoperative assessment of glioma infiltration is essential for maximizing tumor resection while preserving functional brain tissue. Fluorescence lifetime imaging (FLIm) offers real-time, label-free biochemical contrast, but its clinical utility is challenged by biological heterogeneity, class imbalance, and variability in histopathological labeling. We present a data-centric AI (DC-AI) framework that integrates confident learning (CL), class refinement, and targeted label evaluation to develop a robust multi-class FLIm classifier for glioblastoma (GBM) resection margins. FLIm data were collected from 192 tissue margins across 31 newly diagnosed IDH-wildtype GBM patients and initially labeled into seven tumor cellularity classes by an expert neuropathologist. CL was applied to quantify FLIm point-level confidence, identify label inconsistencies, and guide iterative class merging into a three-class scheme ("low", "moderate", "high"). The resulting high-fidelity dataset enabled training a model that achieved 96% accuracy in the three-class task. SHAP analysis revealed class-specific FLIm feature importance, highlighting distinct optical signatures across the infiltration spectrum. Targeted FLIm analysis further identified biological (e.g., gray matter composition) and acquisition-related (e.g., blood contamination) contributors to low-confidence predictions. Blinded re-evaluation of margins flagged by CL demonstrated intra-pathologist variability, underscoring the value of selective relabeling rather than exhaustive review. Together, these findings demonstrate that a DC-AI framework can systematically improve data reliability, enhance model robustness, and refine biological interpretation of FLIm signals, supporting the development of clinically actionable optical tools for real-time glioma margin assessment.

  PublisherDOI

• Label-free fluorescence lifetime imaging can distinguish cancer from healthy tissue in spontaneously occurring canine oral tumors

  Scientific Reports · 2026-01-23

  articleOpen access

  Post-surgical local recurrence of oral cancer remains unacceptably high across species due to the lack of non-invasive tools capable of accurately delineating tumor from healthy tissue. Label-free fluorescence lifetime imaging (FLIm) has shown moderate-to-high success in human head and neck squamous cell carcinoma, but it is unclear whether diagnostic accuracy can be enhanced by incorporating exogenous fluorophores that selectively accumulate in cancer. This study evaluated the performance of 5-ALA induced Protoporphyrin IX (PpIX) fluorescence and autofluorescence FLIm features to discriminate epithelial cancer and healthy tissues in a spontaneous large animal model of disease (15 pet dogs). Fluorescence emission parameters (e.g. lifetimes, intensity ratios, phasors and Laguerre coefficients) differed significantly (p < 0.001) between cancer and healthy tissues in both autofluorescence and PpIX channels. However, autofluorescence features, particularly lifetimes in Channel 1 (390 nm, collagen-sensitive) and intensity ratios in Channel 2 (470 nm, NADH-sensitive), provided the strongest in vivo discrimination. These results demonstrate that label-free FLIm alone is sufficient to distinguish epithelial oral cancers from healthy tissue in dogs, and that the addition of exogenous markers such as 5-ALA–induced PpIX, does not markedly improve diagnostic accuracy enough to warrant incorporation into flourescence imaging approaches.

  PublisherOA PDFDOI

• A Data-Centric Framework for Intraoperative Fluorescence Lifetime Imaging for Glioma Surgical Guidance

  ArXiv.org · 2026-04-28

  articleOpen access

  Accurate intraoperative assessment of glioma infiltration is essential for maximizing tumor resection while preserving functional brain tissue. Fluorescence lifetime imaging (FLIm) offers real-time, label-free biochemical contrast, but its clinical utility is challenged by biological heterogeneity, class imbalance, and variability in histopathological labeling. We present a data-centric AI (DC-AI) framework that integrates confident learning (CL), class refinement, and targeted label evaluation to develop a robust multi-class FLIm classifier for glioblastoma (GBM) resection margins. FLIm data were collected from 192 tissue margins across 31 newly diagnosed IDH-wildtype GBM patients and initially labeled into seven tumor cellularity classes by an expert neuropathologist. CL was applied to quantify FLIm point-level confidence, identify label inconsistencies, and guide iterative class merging into a three-class scheme ("low", "moderate", "high"). The resulting high-fidelity dataset enabled training a model that achieved 96% accuracy in the three-class task. SHAP analysis revealed class-specific FLIm feature importance, highlighting distinct optical signatures across the infiltration spectrum. Targeted FLIm analysis further identified biological (e.g., gray matter composition) and acquisition-related (e.g., blood contamination) contributors to low-confidence predictions. Blinded re-evaluation of margins flagged by CL demonstrated intra-pathologist variability, underscoring the value of selective relabeling rather than exhaustive review. Together, these findings demonstrate that a DC-AI framework can systematically improve data reliability, enhance model robustness, and refine biological interpretation of FLIm signals, supporting the development of clinically actionable optical tools for real-time glioma margin assessment.

  PublisherOA PDF

• Augmenting early stroke diagnosis with an eye-tracker

  Frontiers in Neurology · 2025-10-14

  articleOpen access1st author

  Posterior circulation stroke (PCS) presents significant diagnostic challenges due to poorly localizing and non-specific symptoms, such as dizziness, nausea, and headache, which are often misattributed to benign conditions. This study introduces an innovative diagnostic tool that utilizes a machine learning algorithm-driven eye tracker to enhance early diagnosis of PCS. Our approach involves analyzing eye movements during three standard neurological eye examinations: the Dot Test, H Test, and Optokinetic Nystagmus (OKN) Test. The Discrete Radon Cumulative Distribution Transform (DRCDT) and nearest subspace (NS) classification methods were employed to distinguish between PCS patients and healthy controls by identifying specific eye movement patterns. Results demonstrate that the ensemble model combining the three tests achieved the highest sensitivity and accuracy, with a sensitivity of 96% and an accuracy of 88%, in diagnosing PCS. This study's findings underscore the potential of an eye-tracker-based diagnostic tool to support a more accurate and efficient diagnosis, particularly for non-neurology trained providers, which would improve patient outcomes with more timely and appropriate treatment. The proposed tool offers a practical solution to the limitations of current diagnostic methods, such as the need for calibration and reliance on highly trained specialists, and can be seamlessly integrated into clinical settings to support emergency medical services (EMS) and emergency department (ED) triage.

  PublisherDOI

• Label-free fluorescence lifetime imaging for real-time guidance of stereotactic biopsy: A feasibility study in brain cancer patients

  Neuro-Oncology Advances · 2025-01-01

  articleOpen access

  Background: Stereotactic needle biopsy is frequently indicated for the diagnosis of malignant gliomas in initial management and at tumor recurrence. Current techniques have a sampling error of 24% and diagnostic errors of 10%-30%. Fluorescence lifetime imaging (FLIm) distinguishes among types of brain tissue according to their intrinsic fluorescence properties. Intraprocedural FLIm has been implemented for biopsy guidance. Methods: A FLIm imaging system (fiber-optic probe; pulsed excitation: 355 nm; avalanche photodiode (APD)-based detection: 470/28 nm, 540/50 nm) linked to a stereotactic navigation platform was used to image tissue through the 10-mm side opening of a 1-mm-diameter needle during MRI-directed stereotactic biopsy of 5 suspected malignant gliomas. As the biopsy needle advanced to its targets, FLIm data were continuously acquired and displayed in real time. The data were analyzed by a classification model (trained with an independent dataset from 364 specimens from resection margins of 61 malignant gliomas), and displayed as the probability of the needle's tip lying within the targeted lesion. Predictions from FLIm were compared with biopsy histopathology. Results: Differences in FLIm-derived parameters (intensity, lifetime, Laguerre coefficient) permitted distinguishing lesion (biopsy-confirmed) from uninvolved brain (not biopsy-confirmed) with 81% accuracy and an area under the curve of 88%. Differentiation was better for oligodendroglioma and astrocytoma than for nonenhancing glioblastoma and necrosis. Conclusions: Incorporating FLIm with standard stereotactic biopsy needle and workflow could provide real-time guidance in selecting biopsy targets along the needle's trajectory, potentially improving diagnostic yield and safety. This approach may also aid biopsy of other body tissues and the placement of brain probes treating functional disorders.

  PublisherOA PDFDOI

• How Should We Better Express Respect for Surgical Patients Who Are Incarcerated?

  The AMA Journal of Ethic · 2025-04-01 · 1 citations

  articleOpen access1st authorCorresponding
  PublisherOA PDFDOI

• Data-Centric Learning Framework for Real-Time Detection of Aiming Beam in Fluorescence Lifetime Imaging Guided Surgery

  IEEE Transactions on Biomedical Engineering · 2025-04-15 · 2 citations

  article1st authorCorresponding

  This study introduces a novel data-centric approach to improve real-time surgical guidance using fiber-based fluorescence lifetime imaging (FLIm). A key aspect of the methodology is the accurate detection of the aiming beam, which is essential for localizing points used to map FLIm measurements onto the tissue region within the surgical field. The primary challenge arises from the complex and variable conditions encountered in the surgical environment, particularly in Transoral Robotic Surgery (TORS). Uneven illumination in the surgical field can cause reflections, reduce contrast, and results in inconsistent color representation, further complicating aiming beam detection. To overcome these challenges, an instance segmentation model was developed using a data-centric training strategy that improves accuracy by minimizing label noise and enhancing detection robustness. The model was evaluated on a dataset comprising 40 in vivo surgical videos, demonstrating a median detection rate of 85%. This performance was maintained when the model was integrated in a clinical system, achieving a similar detection rate of 85% during TORS procedures conducted in patients. The system's computational efficiency, measured at approximately 24 frames per second (FPS), was sufficient for real-time surgical guidance. This study enhances the reliability of FLIm-based aiming beam detection in complex surgical environments, advancing the feasibility of real-time, image-guided interventions for improved surgical precision.

  PublisherDOI

• Enhancing Label-Free Fluorescence Lifetime Imaging for Intraoperative Tumor Margin Delineation in Head and Neck Cancer using Data-Centric AI

  Research Square · 2025-05-23 · 1 citations

  preprintOpen access1st authorCorresponding
  PublisherOA PDFDOI

• Early Detection of Lymph Node Metastasis Using Primary Head and Neck Cancer Computed Tomography and Fluorescence Lifetime Imaging

  Diagnostics · 2024-09-23 · 2 citations

  articleOpen access

  Objectives: Early detection and accurate diagnosis of lymph node metastasis (LNM) in head and neck cancer (HNC) are crucial for enhancing patient prognosis and survival rates. Current imaging methods have limitations, necessitating new evaluation of new diagnostic techniques. This study investigates the potential of combining pre-operative CT and intra-operative fluorescence lifetime imaging (FLIm) to enhance LNM prediction in HNC using primary tumor signatures. Methods: CT and FLIm data were collected from 46 HNC patients. A total of 42 FLIm features and 924 CT radiomic features were extracted from the primary tumor site and fused. A support vector machine (SVM) model with a radial basis function kernel was trained to predict LNM. Hyperparameter tuning was conducted using 10-fold nested cross-validation. Prediction performance was evaluated using balanced accuracy (bACC) and the area under the ROC curve (AUC). Results: The model, leveraging combined CT and FLIm features, demonstrated improved testing accuracy (bACC: 0.71, AUC: 0.79) over the CT-only (bACC: 0.58, AUC: 0.67) and FLIm-only (bACC: 0.61, AUC: 0.72) models. Feature selection identified that a subset of 10 FLIm and 10 CT features provided optimal predictive capability. Feature contribution analysis identified high-pass and low-pass wavelet-filtered CT images as well as Laguerre coefficients from FLIm as key predictors. Conclusions: Combining CT and FLIm of the primary tumor improves the prediction of HNC LNM compared to either modality alone. Significance: This study underscores the potential of combining pre-operative radiomics with intra-operative FLIm for more accurate LNM prediction in HNC, offering promise to enhance patient outcomes.

  PublisherOA PDFDOI

• Automatic Recognition of Food Ingestion Environment from the AIM-2 Wearable Sensor

  2024-06-17 · 2 citations

  article

  Detecting an ingestion environment is an important aspect of monitoring dietary intake. It provides insightful information for dietary assessment. However, it is a challenging problem where human-based reviewing can be tedious, and algorithm-based review suffers from data imbalance and perceptual aliasing problems. To address these issues, we propose a neural network-based method with a two-stage training framework that tactfully combines fine-tuning and transfer learning techniques. Our method is evaluated on a newly collected dataset called "UA Free Living Study", which uses an egocentric wearable camera, AIM-2 sensor, to simulate food consumption in free-living conditions. The proposed training framework is applied to common neural network backbones, combined with approaches in the general imbalanced classification field. Experimental results on the collected dataset show that our proposed method for automatic ingestion environment recognition successfully addresses the challenging data imbalance problem in the dataset and achieves a promising overall classification accuracy of 96.63%.

  PublisherDOI

Frequent coauthors

• David Fofi

  Vision pour la Robotique

  18 shared

• Mohamad Naufal Mohamad Saad

  17 shared

• Aamir Saeed Malik

  Boston University

  16 shared

• Gustavo K. Rohde

  University of North Carolina at Chapel Hill

  10 shared

• Yan Zhuang

  National Institutes of Health Clinical Center

  9 shared

• Chad Aldridge

  University of Virginia

  9 shared

• Fabrice Mériaudeau

  Université de Bourgogne

  8 shared

• Andrew M. Southerland

  University of Kentucky

  8 shared

Labs

• Hassan LabPI

Awards & honors

• SPARK at Stanford Translational Science Pilot Grant 2025
• Canadian National STEM Fellowship 2023