Source-linked AI summary
Optimal Control for Cancer Chemotherapy Using Hybrid Quantum Particle Swarm Optimization
Bereket Sitotaw Kidane, Md Samiul Haque Motayed, Shuo Wang
TL;DR
Cancer chemotherapy optimization must account for tumor heterogeneity, mutation dynamics, and limitations of conventional schedules and costate-based methods. The paper combines QPSO global search with regularization-based refinement to produce smooth control trajectories, and its numerical studies report effective periodic and localized dosing strategies for tumor suppression.
Problem
Tumor heterogeneity and mutation dynamics complicate effective chemotherapy, while conventional regimens and PMP-based methods are limited by changing tumor characteristics and sensitivity to initial conditions.
Method
The paper uses QPSO for global dosage exploration followed by fmincon-based regularization refinement to obtain smooth, clinically feasible chemotherapy schedules.
Results
Periodic, especially cosine-based, drug-effectiveness functions support uniform tumor suppression, while combination therapies enhance suppression and reduce toxicity relative to single-drug treatments.
Takeaways & Limitations
Hybrid optimization and periodic dosing provide supported strategies for robust tumor control across heterogeneous tumor traits.
Abstract
from arXiv · showhide
Optimal control in cancer chemotherapy is challenged by tumor heterogeneity and mutations, which complicate treatment effectiveness. Traditional methods, such as Pontryagin's maximum principle (PMP), are often hindered by their reliance on an initial guess for the costate equation, affecting accuracy and convergence. To address these limitations, this work introduces a hybrid quantum particle swarm optimization (QPSO) method based on regularization. QPSO is employed for global exploration to approximate the optimal control trajectory, followed by a regularization-based refinement to ensure smoothness and consistency with optimality conditions. The Hamiltonian function is used for first- and second-order optimality checks, verifying solution quality. Numerical case studies explore various drug effectiveness functions, demonstrating the role of periodic and localized drug delivery in achieving robust tumor suppression and providing insights into the impact of different drug combinations on optimal chemotherapy strategies.
I. INTRODUCTION
Chemotherapy optimization must address heterogeneous, evolving tumors while balancing tumor reduction, toxicity, and clinically feasible dosing. The paper models resistance traits and motivates a hybrid optimization framework to overcome limitations of conventional control methods.
- Tumor heterogeneity and mutation dynamics enable resistant subpopulations to evade treatment, reducing chemotherapy effectiveness.
- Fixed-dose and heuristic chemotherapy schedules often fail to adapt to changing tumor characteristics.
- High-dimensional, nonlinear tumor dynamics make optimization difficult, requiring methods that balance global search, local refinement, and smooth drug administration.
- The model represents drug resistance as a continuous trait x ∈[0, 1], with trait-dependent replication, death, and cytotoxicity parameters.The population density n(t, x) describes cells at resistance level x, and total tumor size is obtained by integrating across traits.
- The objective functional penalizes terminal and cumulative tumor burden together with drug usage or toxicity.For two drugs, the running drug-cost term includes γ1u1(t) + γ2u2(t).
- Mutation effects are modeled through a Gaussian transition kernel and a replication-rate matrix whose off-diagonal terms represent mutation between traits.The mutation fraction θ adjusts replication rates according to mutation probabilities.
III. OPTIMAL CONTROL FORMULATION AND HYBRID QPSO SOLUTION METHOD
The paper formulates chemotherapy planning as a high-dimensional optimal-control problem and introduces a hybrid method combining QPSO with gradient-based fmincon refinement.
- The hybrid optimization approach combines QPSO with gradient-based refinement using fmincon for chemotherapy control under tumor heterogeneity and drug resistance.
A. Step 1: Quantum Particle Swarm Optimization (QPSO)
QPSO is used as the first-stage optimizer to explore dosage schedules globally through quantum-inspired particle updates within drug-concentration bounds.
- QPSO enhances particle-swarm exploration by modeling particles as quantum entities that search the solution space efficiently.Its exploration is guided by local attractors, a contraction-expansion coefficient, random sampling, and a characteristic potential-well length.
- The algorithm initializes particles, personal bests, and a global best, then evaluates fitness and updates these solutions iteratively.
- QPSO terminates when improvement in the global best falls below ϵ and outputs an optimized dosage schedule u∗ QPSO(t).
B. Step 2: Regularization using fmincon
The second stage refines QPSO’s globally searched dosage schedule with fmincon to enforce smoothness and clinical constraints while preserving treatment effectiveness.
- QPSO’s raw control trajectories may contain high-frequency oscillations, so fmincon refines them into smooth, clinically viable schedules.The refinement uses the QPSO schedule as its initial value and enforces 0 ≤u ≤umax.
- The hybrid procedure combines global search with local nonlinear programming to improve chemotherapy-planning robustness and effectiveness.
IV. VERIFICATION: OPTIMALITY CONDITION
The study verifies chemotherapy controls using Hamiltonian first- and second-order conditions. The checks confirm feasibility under control constraints and positive definiteness for Hamiltonian minimization.
- IV. VERIFICATION: OPTIMALITY CONDITION: The first-order condition requires the Hamiltonian gradient with respect to the control to be zero, while the second-order condition verifies minimization.The Hamiltonian formulation includes replication, cytotoxic killing, growth, natural death, control, and costate terms.
- IV. VERIFICATION: OPTIMALITY CONDITION: Constraint-induced deviations of the Hamiltonian gradient from zero occur in some regions because the bounded control trajectory must remain feasible.The control is constrained by u ∈ [0, umax = 3].
V. NUMERICAL EXAMPLES
The numerical examples evaluate double-medication strategies under different drug-effectiveness functions. They assess how modulation strategies affect tumor burden and trait distribution.
- V. NUMERICAL EXAMPLES: Three combination-therapy scenarios compare how different drug-effectiveness modulations influence treatment efficacy.The study seeks to minimize tumor burden while maintaining a uniform distribution across tumor traits.
A. Model Parameters
The model extends cytotoxic-effectiveness functions beyond conventional polynomial or rational forms. It includes localized and periodic functions to represent targeted and non-monotonic drug effects across resistance traits.
- A. Model Parameters: The numerical setup uses 21 equally spaced traits, maximum drug concentrations of 3 for both drugs, mutation parameters σ = 5 and θ = 0.2, and a 10-year simulation.The model also specifies α = 5, β = 400, γ1 = γ2 = 4000, and interaction coefficient ϕ12 = 0.2.
- A. Model Parameters: Localized Gaussian and periodic cosine functions broaden the modeled drug-effectiveness profiles beyond mainly polynomial or rational forms.Gaussian functions represent therapies targeting narrow phenotypes, while cosine functions capture non-monotonic effects across resistance.
B. Combination Therapy Scenarios
The combination-therapy scenarios use complementary drug-effectiveness profiles and hybrid optimization to shape dosage schedules and tumor suppression. The reported dynamics show strong early reductions, later stabilization or resurgence, and trait-selective cytotoxicity.
- 1) Cosine-Gaussian Combination:: In the cosine-Gaussian case, Drug 1 starts at high dosage and declines, while lower-dose Drug 2 complements it by selectively targeting susceptible subpopulations.Tumor burden drops sharply at first but shows a minor later resurgence associated with resistant subpopulations.
- 1) Cosine-Gaussian Combination:: Regularization smooths oscillatory QPSO controls into practical schedules while preserving nearly identical tumor population dynamics.Figure 2 compares raw QPSO controls with refined controls and shows the resulting tumor dynamics and cytotoxic parameters.
- 2) Cosine-Sine Contrast:: The cosine-sine case uses alternating suppression, with Drug 1 tapering from a high dose and Drug 2 remaining sustained to reinforce later treatment.Tumor dynamics sharply decline and then stabilize, with a modest late-stage rise indicating resistant subpopulations.
3) Exponential-Gaussian Mix:
The Exponential–Gaussian regimen pairs an exponentially decaying Drug 1 with a Gaussian Drug 2 to suppress tumors across traits, but lower-trait resistance emerges late.
- 3) Exponential-Gaussian Mix:: The Exponential–Gaussian regimen combines an early, tapering exponential Drug 1 with sustained Gaussian Drug 2 action focused on intermediate traits.Drug 1 targets lower trait values, while Drug 2 concentrates cytotoxic effects on intermediate traits.
- 3) Exponential-Gaussian Mix:: Tumor populations decline sharply at first, then show late-stage resurgence among lower-trait subpopulations, suggesting localized resistance.The killing response decreases monotonically for Drug 1 and follows Gaussian-like modulation for Drug 2.
- 3) Exponential-Gaussian Mix:: Compared with the other strategies, Exponential–Gaussian treatment leaves localized resistance at lower traits, whereas Cosine-Gaussian produces uniform suppression.The comparison is reported for final tumor distributions and temporal population evolution.
VI. DISCUSSION
Drug-effectiveness functions shape how evenly treatment suppresses heterogeneous tumors. Periodic cosine-based combinations provide smoother control, while Exponential–Gaussian treatment leaves localized lower-trait resistance and requires biological-model refinement before clinical use.
- VI. DISCUSSION: Periodic cosine-based functions distribute drug impact more evenly across heterogeneous tumor traits, producing smoother final distributions.Cosine-Gaussian and Cosine-Sine maintain consistent tumor reduction over time.
- VI. DISCUSSION: Cosine-Gaussian achieves uniform suppression, while Cosine-Sine shows intermediate resistance and Exponential-Gaussian leaves localized lower-trait resistance.These differences appear in the final tumor distributions across the three therapy strategies.
- VI. DISCUSSION: Exponential-Gaussian treatment targets specific traits effectively but may allow other subpopulations to persist, particularly at lower trait values.Its temporal dynamics include transient lower-trait resistance and its final distribution is non-uniform.
- VI. DISCUSSION: Combination therapies using periodic cytotoxic functions enhance suppression beyond single-drug treatments while limiting excessive drug exposure.The discussion links this combination strategy to greater control of tumor heterogeneity and reduced overall toxicity.
- VI. DISCUSSION: The mathematical model does not fully capture tumor biology or drug interactions, so parameter tuning and experimental validation remain necessary for clinical application.Future work includes refined resistance and drug-interaction models, adaptive control, and more robust personalized schedules.