Author ORCID Identifier

https://orcid.org/0000-0002-3534-7093

Date of Award

Winter 3-5-2026

Document Type

Thesis (Ph.D.)

Department or Program

Engineering Sciences

First Advisor

Kimberley S. Samkoe

Abstract

The main goal of surgery for head and neck squamous cell carcinoma (HNSCC) is to remove the tumor completely while preserving as much healthy tissue as possible to enhance patient survival and reduce complications. However, achieving clear surgical margins is difficult due to the complex anatomy of the head and neck and the reliance on a surgeon’s visual and tactile assessments. Recently, near-infrared (NIR) fluorescence molecular imaging has shown potential for better tumor visualization during surgery, using targeted imaging agents like epidermal growth factor receptor (EGFR) probes. This has led to the development of fluorescence-guided surgery (FGS), which helps surgeons identify tumor tissue during procedures.

Despite its potential, fluorescence imaging for tumor identification faces challenges such as limited tumor-to-normal contrast and difficulties in quantitatively interpreting fluorescence signals. These challenges stem from physiological factors, like varied delivery of the imaging agents and incomplete tissue penetration, as well as limitations in current image analysis methods. This dissertation explores solutions to these issues, focusing on improving tumor detection and margin assessment in fluorescence imaging for HNSCC.

First, a machine-learning approach was used to enhance tumor classification based on fluorescence imaging characteristics. Analyzing clinical images from ABY-029 Phase-0 trials in HNSCC patients, the new classifiers showed better performance in identifying tumors compared to traditional threshold methods (mean accuracies of 89% vs. 81%, P=0.0072). This advancement allows for more accurate spatial probability mapping of tumor presence in surgical samples.

Second, a framework for standardized fluorescence contrast analysis was created to ensure consistent tumor detection across different clinical datasets. By assessing various contrast metrics and imaging settings, the contrast-to-variance ratio (CVR) emerged as the most effective metric for distinguishing tumors from normal tissue. It demonstrated improved spatial accuracy and reduced sensitivity to imaging variations, making it a reliable basis for assessing surgical margins uniformly across different institutions. The estimated fluorescence margin corresponding to a 5 mm histological margin was found to be 6.16 mm (95% CI: 5.83 - 6.54 mm), confirming that intraoperative fluorescence imaging can serve as a surrogate for assessing pathological margins.

Lastly, preclinical experiments on mice were performed to measure how well EGFR-targeted fluorescence imaging agents penetrate tumor tissue on a microscale. Fluorescence microscopy showed specific penetration patterns and varied distributions over time, influenced by different agents and doses. These observations helped calibrate a transport-binding model that effectively replicated the measured profiles and highlighted the balance between receptor binding, diffusion, and antigen concentration. The model indicated that Panitumumab-IRDye800 achieved its maximum penetration depth of 57 µm after about 40.6 hours, while Cetuximab-IRDye800 reached 55 µm at 36.6 hours, and the affibody imaging agent ABY-029 achieved 57 µm much earlier at around 13.3 hours. The model also suggested optimal times and dosing strategies to enhance tissue penetration. Integration with optical light-propagation simulations illustrated how the microscale distribution of agents affects overall fluorescence contrast, indicating that diagnostic performance depends on both penetration characteristics and vascular distribution.

Overall, these studies create a comprehensive framework that connects molecular delivery mechanisms, fluorescence imaging physics, and quantitative image analysis. This work provides insights for designing better imaging agents, dosing strategies, and analysis methods to enhance tumor identification and margin assessment in FGS for HNSCC.

Download

Share

COinS