Predicting the Progression of Cancerous Tumors in Mice: A Machine and Deep Learning Intuition

Amit K Chattopadhyay,Aimee Pascaline N Unkundiye,Gillian Pearce,Steven Russell

2024-08-01

Abstract:The study explores Artificial Intelligence (AI) powered modeling to predict the evolution of cancer tumor cells in mice under different forms of treatment. The AI models are analyzed against varying ambient and systemic parameters, e.g. drug dosage, volume of the cancer cell mass, and time taken to destroy the cancer cell mass. The data required for the analysis have been synthetically extracted from plots available in both published and unpublished literature (primarily using a Matlab architecture called "Grabit"), that are then statistically standardized around the same baseline for comparison. Three forms of treatment are considered - saline (multiple concentrations used), magnetic nanoparticles (mNPs) and fluorodeoxyglycose iron oxide magnetic nanoparticles (mNP-FDGs) - analyzed using three Machine Learning (ML) algorithms, Decision Tree (DT), Random Forest (RF), Multilinear Regression (MLR), and a Deep Learning (DL) module, the Adaptive Neural Network (ANN). The AI models are trained on 60-80% data, the rest used for validation. Assessed over all three forms of treatment, ANN consistently outperforms other predictive models. Our models predict mNP-FDG as the most potent treatment regime that kills the cancerous tumor completely in ca 13 days from the start of treatment. The models can be generalized to other forms of cancer treatment regimens.

Biological Physics,Soft Condensed Matter

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to predict the progress of cancer cell tumors in mice under different treatment forms through artificial intelligence (AI) technology. Specifically, the research aims to compare the effects of two treatment methods, saline solution and magnetic nanoparticles combined with fluorodeoxyglucose (mNP - FDG), on cancer tumors and predict the time required for complete tumor elimination. The research uses machine - learning (ML) algorithms, such as decision tree (DT), random forest (RF), multiple linear regression (MLR), and a deep - learning (DL) module, namely the adaptive neural network (ANN), to analyze tumor development under different treatment regimens.

The core objective of the paper is to develop a reliable AI toolbox, combining machine - learning and deep - learning models, using data from existing literature to predict results and comparing and validating them with independent experimental results. The research pays special attention to mNP - FDG as the most effective treatment option and predicts that it can completely destroy cancer cell tumors in approximately 13 days.

In addition, the research also explores the impact of different data standardization protocols on the prediction accuracy of the model, and the results show that Min - Max Standardization and Mean Standardization perform best in this regard. Finally, the ANN model performs excellently in all tests, being able to accurately predict tumor progression under saline treatment and also demonstrating the potential of mNP - FDG as a treatment for cancer.

Through these methods, the research shows the powerful ability of AI in predicting cancer tumor growth in mice receiving saline and magnetic nanoparticle treatments, providing new perspectives and tools for future cancer treatment strategies.