James Chun Lam Chow

Agentic artificial intelligence (AI) extends conventional predictive and generative AI by enabling systems to pursue goals through reasoning, planning, information retrieval, memory, tool use, and interaction with clinical workflows. These capabilities create new opportunities for medicine but also a distinct evaluation problem: the reliability of an agent cannot be inferred from the accuracy of its underlying model or final answer alone. Errors may arise during information retrieval, tool use, state updates, inter-agent communication, or autonomous action and may propagate across a clinical workflow. This critical review examines approaches for evaluating and clinically validating agentic AI in medicine. Existing frameworks, including TRIPOD+AI, CONSORT-AI, SPIRIT-AI, DECIDE-AI, STARD-AI, FUTURE-AI, and Good Machine Learning Practice, provide important foundations but were not primarily designed for systems with persistent state, repeated actions, tool use, and multi-agent interaction. Emerging agent benchmarks increasingly address interactive tasks, tool use, trajectory reliability, and adversarial testing. We propose a five-level evaluation framework spanning model, agent, trajectory, system, and clinical outcome, together with a clinical validation ladder progressing from component testing and simulation to retrospective evaluation, prospective silent deployment, supervised clinical use, bounded action, and post-deployment surveillance. Key gaps include standardized trajectory metrics, evaluation of correlated multi-agent errors, autonomy-aware validation, human–agent performance, and lifecycle monitoring. Radiotherapy is used as a high-stakes clinical exemplar. We argue that agentic AI should be evaluated as a dynamic sociotechnical system and that clinical evidence must extend beyond accuracy toward behavioral reliability, safety, and demonstrated clinical benefit.

Precision radiotherapy is becoming increasingly complex, requiring integration of multimodal imaging, treatment planning, adaptive radiotherapy, and diverse clinical data. This paper proposes agentic AI as a responsible clinical decision-support framework in which multiple specialized AI agents collaborate across imaging, contouring, treatment planning, quality assurance, adaptive radiotherapy, response monitoring, and multidisciplinary decision-making. Rather than replacing clinicians, these agents function within a coordinated ecosystem that emphasizes explainability, safety, fairness, privacy, regulatory compliance, and meaningful human oversight. Although challenges remain in clinical validation, interoperability, workflow integration, and governance, agentic AI offers a promising approach to transforming isolated AI applications into transparent, human-centered systems that support personalized cancer treatment.

FLASH radiotherapy, which delivers radiation at ultrahigh dose rates (UHDRs) exceeding 40 Gy/s, has attracted considerable attention because of its potential to spare normal tissues while maintaining tumour control. Gold nanoparticles (GNPs) are promising radiosensitizers that enhance radiation-induced biological effects through increased production of reactive oxygen species (ROS) and subsequent DNA damage. However, the influence of GNPs on DNA damage under FLASH irradiation remains poorly understood. In this study, Geant4-DNA Monte Carlo simulations were performed to investigate the effects of GNP size and dose rate on DNA damage during UHDR electron-beam irradiation. DNA damage was quantified through both direct and indirect mechanisms based on energy deposition in DNA backbone segments and interactions between radiation-induced radical species and DNA, and was evaluated relative to equivalent water nanoparticle (WNP) controls. The results demonstrated a dose rate-dependent reduction in both relative single-strand breaks (RSSBs) and relative double-strand breaks (RDSBs) arising from direct and indirect DNA damage mechanisms. For the 10 keV monoenergetic model condition, the 10 nm GNP produced the greatest radiosensitization among the particle sizes investigated, with the matched GNP/WNP analysis showing a maximum 5-fold enhancement in direct SSB yields. In contrast, substantially weaker GNP-specific enhancement and no comparable size-dependent effect were observed at 1 MeV. These findings demonstrate that GNP-mediated radiosensitization depends on both nanoparticle size and electron energy under the irradiation conditions investigated. The greater enhancement observed for the 10 nm GNP at 10 keV should therefore not be interpreted as identifying a universally optimal GNP size for FLASH radiotherapy.

Background and purpose: The AAPM TG‐43 formalism has long served as the clinical standard for brachytherapy dose calculation but assumes a homogeneous water equivalent medium, overlooking limited scattering conditions and tissue heterogeneities. Model‐based dose‐calculation algorithms (MBDCAs), including Monte Carlo (MC) simulations overcome these limitations by accounting for real tissue composition, scatter, and applicator attenuation. This systematic review evaluates TG‐43, MBDCAs/MC methods in terms of dosimetric accuracy, validation strategies, computational feasibility, clinical implementation barriers, and emerging innovations. Methods: A PRISMA‐guided literature search was conducted using the Scopus database, identifying 284 records, of which 42 full‐text studies met inclusion criteria. Eligible studies compared at least two of the three dose‐calculation approaches (TG‐43, MBDCAs and MCs) in pelvic, breast, or head‐and‐neck brachytherapy. Extracted data encompassed dosimetric discrepancies, validation approaches, computational performance, workflow integration, and enabling technologies. Results: Across anatomical sites, TG‑43 showed no consistent bias. Its differences from heterogeneity‑aware models depended on tissue composition, scatter conditions, and source geometry. In soft‑tissue regions with minimal heterogeneity, TG‑43 generally overestimated target coverage by about 0.5%–5%. Near low‑density interfaces or in reduced‑scatter configurations, TG‑43 could instead yield lower doses than Monte Carlo or MBDCAs. For OARs, discrepancies were site‑specific: skin dose was often overestimated, while other organs showed smaller or opposite variations. Overall, MBDCAs and Monte Carlo agreed with experimental or benchmark data within roughly 3% and produced more reliable biological metrics. GPU‑accelerated and deep‑learning engines reduced computation times from hours to seconds, shifting remaining challenges toward standardization, commissioning, and QA. Successful clinical adoption relied on TG‑186–aligned validation, staff training, and integrated automated workflows. Conclusion: Evidence from the systematic review supports a clinical transition toward TG‐186–compliant, heterogeneity‐aware dose‐calculation frameworks. MBDCAs/MC algorithms provide superior dosimetric and radiobiological accuracy and are increasingly compatible with adaptive and biologically guided planning. Broad implementation is now supported by established QA standards and benchmarking datasets, which connects dosimetric precision with patient outcomes. The transition beyond TG‐43 marks a pivotal step toward precision, safety, and personalization in modern brachytherapy.

Background and Objective: Radiation treatment planning is a complex and resource-intensive process that plays a critical role in the quality and safety of radiotherapy. Increasing treatment complexity and growing demands on clinical workflows have motivated the adoption of artificial intelligence (AI) to support planning tasks. While early AI applications focused primarily on automation, recent developments have expanded their role toward clinical decision support. The objective of this review is to provide a comprehensive overview of current AI applications across the treatment planning workflow and to introduce a conceptual framework that characterizes the transition from task-based automation to clinically integrated decision support. Methods: This narrative review summarizes current applications of AI in radiation treatment planning, organized by key stages of the planning workflow. We examine developments in auto-segmentation, dose prediction, automated plan generation, and plan evaluation, with a focus on clinical evidence, translational readiness, and implementation challenges. Key Content and Findings: AI-assisted contouring and knowledge-based planning represent the most mature applications and are increasingly integrated into clinical practice. Emerging AI tools support plan evaluation and optimization by providing benchmarks, exploring trade-offs, and promoting consistency across planners and institutions. Rather than replacing human expertise, these systems function most effectively as decision-support tools that augment clinical judgment. However, challenges related to data quality, generalizability, interpretability, and safety continue to limit widespread adoption. Conclusions: AI is reshaping radiation treatment planning, with a clear shift from task-level automation toward clinically integrated decision support. Responsible translation requires rigorous validation, human-centered design, and attention to equity and safety. When implemented thoughtfully, AI has the potential to enhance decision-making, reduce variability, and improve the quality and consistency of radiotherapy care.

Tumorigenesis and therapeutic resistance [1][2][3]. Several contributions in this collection build on this conceptual foundation to deepen the mechanistic understanding of cancer development and treatment response.Brahme [4] presents a comprehensive review of quantum scale radiation interactions describing how low-energy delta rays can produce concentrated DNA damage that influences biological effectiveness and informs future optimization of fractionation strategies and linear energy transfer selection in clinical radiotherapy. Complementing this perspective, Liu et al. [5] explore how RNA methylation marks such as m6A, m5C, and m7G regulate gene expression and contribute to radioresistance, outlining potential strategies for targeting these pathways to increase therapeutic sensitivity. Hormone signaling and receptor conformation also feature prominently in this collection. Saha and Lukong [6] provide a detailed structural and functional overview of ERα signaling, ESR1 mutations, and GPER activity in endocrine-resistant breast cancer and highlight the need for therapeutic strategies that address both nuclear and membrane mediated signaling. Moreover, Ji et al. [7] expanded the biomarker landscape by reviewing claudin family proteins, describing their roles as modulators of cell adhesion, invasion, and metastasis, and discussing the emergence of claudin targeted therapies across multiple cancer types.Other mechanistic work focuses on pathways not traditionally centered in oncology.Meyer et al. [8] examine the role of the coagulation factors FXa and PAR2 in colon cancer progression and reveal how FXa signaling enhances tumor proliferation and migration, providing a novel axis for potential intervention. Akhtar et al. [9] study several microRNAs in leiomyosarcoma and confirm their upregulation in tumor tissue while noting their limited predictive value for metastatic risk, which is important for future biomarker validation. In addition, Ma et al. [10] address a long standing clinical problem by reviewing high risk contributors to radioiodine refractoriness in differentiated thyroid cancer. Their analysis includes demographic factors, genetic mutations such as BRAF and RAS, histopathological patterns, serum markers, and comorbidity related influences on iodine uptake. These papers collectively show how a deeper molecular understanding provides the foundation for more precisely targeted and biologically informed therapies.Technological innovation continues to reshape oncology by translating biological knowledge into improved diagnostic and therapeutic approaches. Large-scale cancer genomics initiatives such as The Cancer Genome Atlas and the International Cancer Genome Consortium have revealed the genomic heterogeneity of tumors and identified actionable molecular alterations across numerous cancer types [11][12][13]. These advances underpin the emergence of precision oncology and increasingly data-driven clinical decision-making.Several contributions highlight how evolving technologies and analytical methods are reshaping treatment design and clinical decision making. Amin et al. [14] analyze survival outcomes in more than 117,000 sarcoma patients in the United States National Cancer Database and report that compared with photon therapy, proton beam therapy is associated with improved overall survival in several clinically important histologies.Sun et al. [15] review international experience with boron neutron capture therapy for melanoma and outline recent improvements in accelerator systems and boron delivery agents that support continued development. On the other hand, Aigner et al. [16] introduce a regional treatment approach for anal squamous cell carcinoma that uses isolated hypoxic pelvic perfusion with or without reversible electroporation. Their reported cases show complete clinical and pathological remissions with minimal toxicity, demonstrating a promising alternative for select patients.Precision oncology also benefits from advances in molecular profiling. Erices et al. [17] produce an ancestry informed genomic map of gallbladder cancer in Chile, identifying actionable alterations in genes including ATM, BRCA1/2, EGFR, and ERBB2. They also report an association between Mapuche ancestry and TP53 mutations, underscoring the value of population specific genomic evidence. Hofmann et al. [18] present a case of colorectal cancer with an ALK CEP44 fusion that initially responds to ALK inhibition before the rapid development of multiple resistance mutations, illustrating the dynamic nature of tumor evolution and the need for serial molecular assessment.Machine learning is represented by Antysheva et al. [19] who develop a single sample molecular classifier for cancer grading that operates without batch correction and provides reliable risk stratification for intermediate grade tumors, an important step toward real world applicability. Important insights into surgical practice and outcome prediction are also included in this group. Yumak and Demir [20] report that lymphovascular invasion, perineural invasion, poor differentiation, R1 resection, emergency surgery, and subsequent metastasis are dominant predictors of mortality in patients with rectal cancer. Furthermore, Lin et al. [21] evaluate risk factors for contralateral central lymph node metastasis in papillary thyroid cancer and show that microcalcification and Hashimoto thyroiditis are strong predictors that support selective contralateral dissection. These papers collectively demonstrate how technologies such as particle therapy, machine learning, advanced profiling, and carefully analyzed clinical data are transforming how clinicians tailor treatment for specific patients and settings.Rare cancers collectively represent a substantial yet often underrecognized portion of the global cancer burden, accounting for approximately 20-25% of malignancies [22,23]. Their study provides unique opportunities to uncover novel...

The advancement of generative artificial intelligence (GAI) in healthcare is driving a transition from dialogue-based medical chatbots to workflow-embedded clinical AI agents. These agentic systems incorporate persistent state management, coordinated tool invocation, and bounded autonomy, enabling multi-step reasoning within institutional processes. As a result, traditional response-level evaluation frameworks are insufficient for understanding system behavior. This review provides a conceptual synthesis of the evolution from conversational systems to agentic architectures and proposes a system-level modeling framework for ethical clinical AI agents. We identify core architectural dimensions, including autonomy gradients, state persistence, tool orchestration, workflow coupling, and human–AI co-agency, and examine how these features reshape bias propagation pathways, error cascade dynamics, trust calibration, and accountability structures. Emphasizing that ethical risks emerge from longitudinal system interactions rather than isolated outputs, we argue for embedding fairness constraints, transparency mechanisms, and lifecycle governance directly within AI design. By outlining trajectory-level evaluation strategies, equity-aware development approaches, collaborative oversight models, and adaptive regulatory frameworks, this paper establishes a foundation for the responsible and trustworthy integration of agentic AI in healthcare.

Machine learning (ML) is transforming cancer imaging by enhancing diagnosis, automating image analysis, and improving treatment planning. This review explores key ML applications, including tumor detection, radiomics, multi-modal imaging, and therapy monitoring. We discuss fundamental ML techniques, deep learning architectures, and data preprocessing strategies essential for medical imaging. ML-driven approaches have improved tumor segmentation, feature extraction, and computer-aided diagnosis across various cancer types. In cancer therapy, artificial intelligence (AI) aids radiotherapy planning, treatment response prediction, and real-time image-guided interventions. However, challenges such as data scarcity, model bias, and regulatory hurdles limit clinical adoption. Emerging solutions include explainable AI, federated learning for data privacy, and quantum computing for advanced imaging. Addressing these challenges through interdisciplinary collaboration will accelerate AI integration into clinical practice, enhancing cancer diagnosis and treatment.