Hybrid Machine Learning and Mathematical Modeling for Tumor Dynamics Prediction: Comparing SPIONs against mNP-FDG
∗ Amit K Chattopadhyay, Aimee Pascaline N Unkundiye, Gillian Pearce · 2025
Read Hybrid Machine Learning and Mathematical Modeling for Tumor Dynamics Prediction: Comparing SPIONs against mNP-FDG by ∗ Amit K Chattopadhyay, Aimee Pascaline N Unkundiye, Gillian Pearce (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-21094.txt.
Introduction
Amit K Chattopadhyay, Aimee Pascaline N Unkundiye, and Gillian Pearce blend machine learning with math models to forecast how a tumor will grow under two nanoparticle treatments. One treatment uses sugar-coated particles that home in on cancer cells. The other uses iron oxide particles that respond to magnets. Both aim at zone-specific radiation therapy.
The models tell a split story. The sugar-coated particles tamed tumor growth within two days, far faster than the iron particles' eighteen. Yet the iron particles wiped the tumor out in twenty days, while the sugar-coated ones needed more than forty. The authors built a small interactive tool for trying treatment choices, and they suggest a sequence: first slow the spread, then finish the tumor off.
Catalog record
| CREATOR | ∗ Amit K Chattopadhyay, Aimee Pascaline N Unkundiye, Gillian Pearce |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-21094
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hybrid Machine Learning and Mathematical Modeling for Tumor Dynamics Prediction: Comparing SPIONs against mNP-FDG ∗ Amit K Chattopadhyay, Aimee Pascaline N Unkundiye, Gillian Pearce Abstract Outline : Since the late eighties, Superparamagnetic Iron Oxide Nanoparticles (SPIONs) have been extensively studied for their exceptional ability in targeted drug delivery, typically coerced by magnetic fields and delivered at chemotherapy sites. Both iron-coated (SPIONs) and fluorodeoxyglucose-coated (mNP-FDG) magnetic nanoparticles are known to be highly potent in their ability to deliver drugs with uncanny precision to cancerous cells while minimizing damage to healthy cells. Of late, though, questions have been raised about the potential increase in cytotoxicity, particularly with SPIONs. Combining Machine Learning (Extreme Gradient Boosting) with continuum modeling (exponential and logistic growth), we find that while mNP-FDG can control tumor progression within 2 days compared to 18 days by SPIONs, for complete termination of the tumor, SPIONS (20 days) are superior compared to mNP-FDG (more than 40 days). We also provide an interactive graphical user interface (GUI) developed with Tkinter/Python that allows users to input relevant data, such as treatment type and time, to receive real-time tumor volume predictions. Our ML-guided prediction indicates joint therapy as the optimum choice, with mNP-FDG ideal for taming the tumor spread, followed by SPIONs for complete eradication, facilitating personalized cancer treatment in clinical practice. Main Limitations : This modeling study uses trendline data from multiple published sources that may have been conducted under varying experimental conditions. Thus, data optimization and normalization is a potential challenge. Also, different forms of cancer may have been addressed. However, our robust modeling infrastructure ensures genericity through multiple ensemble averaging of results that consistently converge. Objective: The present study aims at a comparative analysis of two groups of inorganic molecules, Superparamagnetic Iron Oxide Nanoparticles (SPIONs) and fluorodeoxyglucose-coated (mNP-FDG) magnetic nanoparticles, to understand the relative merits and demerits in optimizing specificity and cytotoxicity of chemotherapy treatment when administered through these agents. We also analyze the possibility of combinatorial drug administration using both agents. Keywords: Tumor dynamics; SPION; mNP-FDG; Machine Learning; AI; GUI. I Introduction Cancer remains one of the leading causes of mortality worldwide, with an estimated 9.6 million deaths annually WHO2020 . Despite significant advancements in chemotherapy, radiation therapy, and immunotherapy, significant challenges persist, including drug resistance, off-target effects, and limited precision in treatment delivery. Nanotechnology has emerged as a promising avenue to address these challenges by enabling targeted drug delivery and improving treatment efficacy. Among the various nanocarrier systems, superparamagnetic iron oxide nanoparticles (SPIONs) and fluorodeoxyglucose-coated magnetic nanoparticles (mNP-FDG) represent two innovative strategies with considerable potential in oncological applications. SPIONs have been extensively investigated for their multifunctional applications in cancer therapy, including drug delivery and antibody conjugation . However, their primary therapeutic potential lies in their ability to facilitate magnetic hyperthermia, wherein exposure to an alternating magnetic field induces localized heating, leading to cancer cell apoptosis Russell2021 . This property has positioned SPIONs as an attractive candidate for integration with MR-Linac technology, which combines magnetic resonance imaging (MRI) with radiotherapy to enhance tumor targeting. Iron oxide contrast agents, traditionally employed in T2-weighted MRI scans, may also enhance radiosensitivity, as demonstrated by gadolinium-based nanoparticles such as AGuIX . Previous studies have explored the radiosensitizing effects of SPIONs with Russell et al reporting enhanced cytotoxicity and reactive oxygen species (ROS) production in MCF7 breast cancer cells while Kirakli et al. investigated citrate-coated SPIONs for similar applications Kirakli2018 . In contrast, mNP-FDG exploits the metabolic characteristics of cancer cells by mimicking glucose uptake pathways, thereby enabling high-specificity tumor targeting Aras2018 ; Chattopadhyay2024 . Cancer cells exhibit an increased reliance on glycolysis for energy production, a phenomenon known as the Warburg effect. By leveraging this metabolic adaptation, mNP-FDG can be selectively internalized by malignant cells, potentially increasing therapeutic efficacy while minimizing damage to healthy tissues. Additionally, mNP-FDG nanoparticles offer the potential for dual diagnostic and therapeutic applications, as they can serve as contrast agents for positron emission tomography (PET) imaging while concurrently delivering therapeutic payloads. However, key challenges remain, including optimizing their stability, mitigating cytotoxicity, and assessing their efficacy across various cancer subtypes. Recent studies have demonstrated the potential of mNP-FDG in preclinical models, yet further investigation is required to fully elucidate their mechanisms of action and therapeutic potential in clinical settings Vangijzegem2023 . In recent years, the integration of computational modeling and artificial intelligence (AI) has opened new avenues for optimizing nanoparticle based therapies. In this study, mathematical modeling provides a framework for understanding tumor growth dynamics and drug interactions, while machine learning (ML) algorithms enable the identification of complex, nonlinear patterns within experimental datasets. By combining these methodologies, researchers aim to achieve more accurate predictions, optimize treatment protocols, and advance personalized cancer therapy. In this study, we propose a computational framework that compares mathematical modeling with machine learning (ML) to predict tumor dynamics under SPION and mNP-FDG treatments. Mathematical models provide mechanistic insights into tumor growth and drug interactions, while ML models capture nonlinear relationships and patterns in complex datasets. By integrating these approaches, we aim to enhance prediction accuracy, optimize treatment strategies, and contribute to the development for personalized cancer therapy. II Methodology II.1 Data Collection and Preprocessing Tumor volume data were acquired from experimental and clinical studies involving SPIONs and mNP-FDG as therapeutic agents Vangijzegem2023 ; Chattopadhyay2024 . These datasets provided detailed measurements of tumor volumes, treatment regimens, and their temporal evolution over specific intervals. Data extraction was performed using MATLAB’s Grabit tool, which facilitated digitization from figures from published literature. The tool was initially tested to extract mNP-FDG data from Gao2015 ; Qi2022 , followed by implementation on the SPION database. Preprocessing was not required, as the extracted data was already in a consistent analyzable format. II.2 Machine Learning Integration Since the experiments were conducted by labeling the samples with the specific therapeutic agent, i. e. mNP-FDG or SPION, supervised machine learning models were implemented. We chose ensemble techniques, specifically random forests and gradient boosting algorithms. The choice was largely guided by the fact that ensemble supervised learning methods enhance predictive performance by combining multiple models, reducing bias, variance, and overfitting while improving generalization. By leveraging diverse learning perspectives, they capture complex patterns and make predictions more stable and robust. These methods effectively utilize weak learners as would be the case for our data that were both sparse and also mixed with other descriptors, thus reducing the risk of poor model selection and enhancing adaptability to various data distributions. The input features from the data targeted initial tumor volume, treatment type (SPIONs or mNP-FDG), and time points (longitudinal), while the target variable was tumor volume at subsequent time intervals. Model performance was evaluated using cross-validation, ensuring robustness and generalizability of predictions. II.3 Development of a Graphical User Interface (GUI) To enable direct usage of this algorithm in clinical studies, an interactive GUI was developed using Python’s Tkinter library to enable real-time predictions of tumor dynamics from the longitudinal data. The GUI allows users to input treatment parameters such as therapy type and time elapsed offering a visual representation of predicted tumor growth trajectories under various therapeutic scenarios. This user-friendly interface aims to enhance clinical decision-making by streamlining the evaluation of treatment strategies. III Results III.1 Efficacy of SPIONs and mNP-FDG