CoChat
Sign Up
Cited by 7Artificial Intelligence in Healthcare and EducationAI in cancer detectionRadiomics and Machine Learning in Medical Imaging

Artificial Intelligence in Clinical Oncology: From Productivity Enhancement to Creative Discovery

Open access
Kuno, Masahiro, Osumi, Hiroki, Udagawa, Shohei, Yoshikawa, Kaoru, Ooki, Akira, Shinozaki, Eiji, Ishikawa, Tetsuo, Oba, Junna +2 more·2025·Current Oncology
DOI 10.3390/curroncol32110588PMC PMC12651923
Open this paper in CoChat — free accountOpen PDF

Abstract

Modern clinical oncology faces an unprecedented data complexity that exceeds human analytical capacity, making artificial intelligence (AI) integration essential rather than optional. This review examines the dual impact of AI on productivity enhancement and creative discovery in cancer care. We trace the evolution from traditional machine learning to deep learning and transformer-based foundation models, analyzing their clinical applications. AI enhances productivity by automating diagnostic tasks, streamlining documentation, and accelerating research workflows across imaging modalities and clinical data processing. More importantly, AI enables creative discovery by integrating multimodal data to identify computational biomarkers, performing unsupervised phenotyping to reveal hidden patient subgroups, and accelerating drug development. Finally, we introduce the FUTURE-AI framework, outlining the essential requirements for translating AI models into clinical practice. This ensures the responsible deployment of AI, which augments rather than replaces clinical judgment, while maintaining patient-centered care.

Verified 2026-10-07 · 6 references not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

This narrative review argues that AI in clinical oncology improves productivity through automation and enables creative discovery via multimodal computational biomarkers, unsupervised phenotyping, and drug discovery. It introduces the FUTURE-AI framework as a guide for fair, reliable, and explainable clinical translation.

Key findings

  • AI productivity gains include FDA clearance/approval of ~700 AI/ML-enabled medical devices by 2024, 76% related to radiology; in the MASAI trial, AI-supported mammography screening across 105,934 women was non-inferior and reduced radiologist workload by 44.2% (61,248 vs. 109,692 reads).
  • Diagnostic AI performance is substantial but variable: a meta-analysis of 100 pathology studies with >152,000 whole-slide images reported mean sensitivity 96.3% and specificity 93.3%, with breast lower (83%/88%) than gastrointestinal (93%/94%) or urologic (95%/96%); colonoscopy AI improved adenoma detection rate by ~14% relative.
  • Creative discovery examples include TOAD, trained on 17,486 slides across 18 primary sites, achieving top-1/top-3 accuracy of ~0.83/0.96 internally and 0.80/0.93 externally, and matching pathologist differentials in 82% of a 317-patient cancer-of-unknown-primary cohort; SCORPIO covered nearly 10,000 ICI-treated patients across >20 cancers and outperformed TMB and PD-L1 for survival/benefit prediction.
  • Drug-discovery examples include an AlphaFold2/generative pipeline for hepatocellular carcinoma that produced ~9,000 candidate molecules, with 7 synthesized, 1 lead, in ~30 days; FUTURE-AI was developed by 117 experts from 50 countries, yet a systematic review found a mean FUTURE-AI score of only 5.1/30.

Methods

  • Narrative review tracing AI evolution in oncology from traditional machine learning in the 1990s to deep learning in the 2010s and transformer-based foundation models since 2017.
  • Synthesizes published clinical applications across radiology, digital pathology, endoscopy, dermatology, NLP/LLMs, computational biomarkers, unsupervised phenotyping, and drug discovery.
  • Introduces and applies the FUTURE-AI framework’s six principles—fairness, universality, traceability, usability, robustness, and explainability—to clinical translation; no primary data collection or systematic search protocol is described.

Limitations

  • Stated: AlphaFold predicts a protein’s most stable shape but does not model protein motion, hidden pockets, or complex changes; LLM hallucination remains a safety concern, and RAG reduces but does not eliminate it; external validation of Woollie dropped AUROC from 0.97 to 0.88, indicating generalization challenges.
  • Stated: regulatory frameworks for AI in oncology remain evolving, with gaps in liability assignment when AI-assisted decisions cause harm; comprehensive evaluation of radiology report-generation time savings, including verification and correction, remains limited.
  • Evident from the text: as a narrative review, it does not report a systematic search strategy, risk-of-bias assessment, or meta-analysis, and its performance examples are drawn from cited studies rather than independently validated in a single dataset.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/curroncol32110588

Cites

152 of 158 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 154 cited sources checked · 0 supported · 0 flagged

  • Premisereasoning

    Deep learning has advanced image analysis in oncology in fields such as dermatology, radiology, endoscopy, and pathology.

    “In oncology, DL has advanced image analysis in fields such as dermatology, radiology, endoscopy, and pathology.”
    Quote in the paperNot a substantial claim
  • Premisecitation

    In the MASAI trial, AI-supported screening reduced radiologist workload by 44.2%.

    “the AI-supported screening was non-inferior and reduced radiologist workload by 44.2% (AI arm, 61,248 reads; control, 109,692) without increasing false-positive rates, and with evidence of earlier cancer detection [30,31].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [30] Kristina, 2023, Artificial Intelligence-Supported Screen Reading versus Standard Double Reading in the Mammography Screening with Artificial Intelligence Trial (MASAI): A Clinical Safety Analysis of a Randomised, Controlled, Non-Inferiority, Single-Blinded, Screening Accuracy Study, Lancet Oncol. source

      Not checked in the free run.

    • cites [31] Veronica, 2025, Screening Performance and Characteristics of Breast Cancer Detected in the Mammography Screening with Artificial Intelligence Trial (MASAI): A Randomised, Controlled, Parallel-Group, Non-Inferiority, Single-Blinded, Screening Accuracy Study, Lancet Digit. Health source

      Not checked in the free run.

  • Premisecitation

    In the MASAI trial, AI-supported screening was non-inferior to standard double reading.

    “the AI-supported screening was non-inferior and reduced radiologist workload by 44.2% (AI arm, 61,248 reads; control, 109,692) without increasing false-positive rates, and with evidence of earlier cancer detection [30,31].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [30] Kristina, 2023, Artificial Intelligence-Supported Screen Reading versus Standard Double Reading in the Mammography Screening with Artificial Intelligence Trial (MASAI): A Clinical Safety Analysis of a Randomised, Controlled, Non-Inferiority, Single-Blinded, Screening Accuracy Study, Lancet Oncol. source

      Not checked in the free run.

    • cites [31] Veronica, 2025, Screening Performance and Characteristics of Breast Cancer Detected in the Mammography Screening with Artificial Intelligence Trial (MASAI): A Randomised, Controlled, Parallel-Group, Non-Inferiority, Single-Blinded, Screening Accuracy Study, Lancet Digit. Health source

      Not checked in the free run.

  • Premisecitation

    Traditional machine learning models have supported tumor classification, survival prediction, and risk stratification using structured clinical or genomic data.

    “traditional ML models, such as support vector machines (SVM), decision trees, and random forests, have supported tumor classification, survival prediction, and risk stratification using structured clinical or genomic data [6,12,13,14].”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [6] Beam, 2023, Artificial Intelligence in Medicine, N. Engl. J. Med. source

      Not checked in the free run.

    • cites [12] Cruz, 2007, Applications of Machine Learning in Cancer Prediction and Prognosis, Cancer Inform.

      Not checked in the free run.

    • cites [13] Kourou, 2015, Machine Learning Applications in Cancer Prognosis and Prediction, Comput. Struct. Biotechnol. J. source

      Not checked in the free run.

    • cites [14] Tseng, 2020, Machine Learning and Imaging Informatics in Oncology, Oncology source

      Not checked in the free run.

  • Premisenone given

    In mammography, deep learning systems have achieved radiologist-comparable breast cancer detection.

    “In mammography, deep learning systems have achieved radiologist-comparable breast cancer detection.”
    Quote in the paperSubstantial claim
  • May combine two claims. 46
  • Classified as probably not a substantial claim. 10

  • May combine two claims.

    Premisecitation

    Transformer-based methods support clinical text processing, whole-slide and radiologic image interpretation, and genomic sequence modeling in oncology.

    “In oncology, transformer-based methods support clinical text processing, whole-slide and radiologic image interpretation (vision transformers), and genomic sequence modeling [24,25,26,27].”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [24] Pai, 2024, Foundation Model for Cancer Imaging Biomarkers, Nat. Mach. Intell. source

      Not checked in the free run.

    • cites [25] Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., and Gelly, S. (2020). An Image Is Worth 16x16 Words: Transformers for Image Recognition at Scale. arXiv.

      Not checked in the free run.

    • cites [26] Yan, 2025, A Multimodal Vision Foundation Model for Clinical Dermatology, Nat. Med. source

      Not checked in the free run.

    • cites [27] Cui, 2024, scGPT: Toward Building a Foundation Model for Single-Cell Multi-Omics Using Generative AI, Nat. Methods source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    As of 2024, the FDA had cleared or approved approximately 700 AI/ML-enabled medical devices, 76% of which are related to radiology.

    “As of 2024, the FDA had cleared or approved approximately 700 AI/ML-enabled medical devices, 76% of which are related to radiology.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [1] Lotter, 2024, Artificial Intelligence in Oncology: Current Landscape, Challenges, and Future Directions, Cancer Discov. source

      Not checked in the free run.

    • cites [29] Panagiotis, 2025, Implementing Artificial Intelligence Algorithms in the Radiology Workflow: Challenges and Considerations, Mayo Clin. Proc. Digit. Health source

      Not checked in the free run.

  • May combine two claims.

    Premisenone given

    A key limitation of traditional machine learning is the need for manual feature engineering, which requires domain expertise and limits scalability for unstructured inputs.

    “However, a key limitation is the need for manual feature engineering, which requires domain expertise and limits scalability for unstructured inputs.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    AI integration in clinical oncology is essential rather than optional.

    “making artificial intelligence (AI) integration essential rather than optional”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    AI enables creative discovery in cancer care by integrating multimodal data to identify computational biomarkers, performing unsupervised phenotyping to reveal hidden patient subgroups, and accelerating drug development.

    “AI enables creative discovery by integrating multimodal data to identify computational biomarkers, performing unsupervised phenotyping to reveal hidden patient subgroups, and accelerating drug development.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    AI in oncology has evolved from traditional machine learning in the 1990s to deep learning in the 2010s and, since 2017, to transformer-based foundation models.

    “AI has progressed from traditional machine learning (ML) in the 1990s to deep learning (DL) in the 2010s and, since 2017, to transformer-based foundation models.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    AI enhances productivity in cancer care by automating diagnostic tasks, streamlining documentation, and accelerating research workflows.

    “AI enhances productivity by automating diagnostic tasks, streamlining documentation, and accelerating research workflows across imaging modalities and clinical data processing.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    AI systems for clinical trial operations can parse biomarker requirements, prior therapies, and performance status from EHRs, identify potentially eligible patients, and flag candidates for screening.

    “Systems parse biomarker requirements, prior therapies, and performance status from EHRs, identify potentially eligible patients, and flag candidates for screening [56,57,58,59,60,61,62].”
    Quote in the paperSubstantial claim0 of 7 sources checked
    • cites [56] Wornow, 2025, Zero-Shot Clinical Trial Patient Matching with LLMs, NEJM AI source

      Not checked in the free run.

    • cites [57] Hullings, 2021, Application of Information Technology to Clinical Trial Evaluation and Enrollment: A Review: A Review, JAMA Oncol. source

      Not checked in the free run.

    • cites [58] Chow, 2023, Use of Artificial Intelligence for Cancer Clinical Trial Enrollment: A Systematic Review and Meta-Analysis, J. Natl. Cancer Inst. source

      Not checked in the free run.

    • cites [59] Beck, 2020, Artificial Intelligence Tool for Optimizing Eligibility Screening for Clinical Trials in a Large Community Cancer Center, JCO Clin. Cancer Inform. source

      Not checked in the free run.

    • cites [60] Lu, 2024, Artificial Intelligence for Optimizing Recruitment and Retention in Clinical Trials: A Scoping Review, J. Am. Med. Inform. Assoc. JAMIA source

      Not checked in the free run.

    • cites [61] Saady, 2025, Implementation of Artificial Intelligence Approaches in Oncology Clinical Trials: A Systematic Review, Artif. Intell. Med. source

      Not checked in the free run.

    • cites [62] Homolak, 2023, Opportunities and Risks of ChatGPT in Medicine, Science, and Academic Publishing: A Modern Promethean Dilemma, Croat. Med. J. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    AI's greater impact in oncology may be its ability to enable new discoveries.

    “While productivity gains are important, AI’s greater impact in oncology may be its ability to enable new discoveries.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    This automation reduces reporting variability and improves consistency in evaluating treatment responses across different radiologists and institutions.

    “This automation reduces reporting variability and improves consistency in evaluating treatment responses across different radiologists and institutions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    AI improves research productivity by making literature searches more efficient and accelerating data analysis, allowing investigators to spend more time on hypothesis generation and interpretation.

    “AI improves research productivity by making literature searches more efficient and by accelerating data analysis, allowing investigators to spend more time on hypothesis generation and interpretation rather than manual screening and coding.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisereasoning

    In evaluations of lung and gastroesophageal cancers, MUSK outperformed the standard PD-L1 biomarker alone in predicting immunotherapy benefit and surpassed clinicopathologic baselines for survival-related endpoints.

    “In evaluations of lung and gastroesophageal cancers, MUSK outperformed the standard programmed cell death ligand 1 (PD-L1) biomarker alone in predicting immunotherapy benefit and surpassed clinicopathologic baselines for survival-related endpoints.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Prov-GigaPath was pretrained on more than 170,000 slides and is among the first open-access foundation models for digital pathology.

    “pretrained Prov-GigaPath on more than 170,000 slides (>1 billion 256 × 256-pixel tiles) and is among the first open-access foundation models for digital pathology”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [36] Xu, 2024, A Whole-Slide Foundation Model for Digital Pathology from Real-World Data, Nature source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    The performance drop of Woollie from internal to external validation indicates challenges in generalization across institutions, but the model maintains clinically relevant predictive capability.

    “While this performance drop indicates challenges in generalization across institutions, the model maintains clinically relevant predictive capability, demonstrating the importance of diverse training data and rigorous external validation.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Vibe coding carries substantial risks when users lack fundamental statistical and analytical knowledge.

    “While vibe coding offers remarkable convenience, it carries substantial risks when users lack fundamental statistical and analytical knowledge.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    TOAD achieved top-1/top-3 accuracies of approximately 0.83/0.96 on an internal test set and 0.80/0.93 on an external test set; in a CUP cohort (n=317), its top-three predictions matched pathologists' differentials in 82% of cases.

    “Trained on 17,486 slides spanning 18 primary sites, it achieved top-1/top-3 accuracies of approximately 0.83/0.96 on an internal test set and 0.80/0.93 on an external test set; in a curated cancer of unknown primary (CUP) cohort (n = 317), its top-three predictions matched pathologists’ differentials in 82% of cases.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [82] Nangalia, 2019, Genome Sequencing during a Patient’s Journey through Cancer, N. Engl. J. Med. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Prov-GigaPath has shown strong performance in cancer classification, mutation prediction, and vision-language tasks.

    “it has shown strong performance in cancer classification, mutation prediction, and vision-language tasks [36].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [36] Xu, 2024, A Whole-Slide Foundation Model for Digital Pathology from Real-World Data, Nature source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Generative AI systems for radiology report drafting can generate structured reports, standardize terminology, and guarantee inclusion of critical elements such as tumor measurements and response assessments according to RECIST guidelines.

    “These systems can generate structured reports based on imaging findings, standardize terminology, and guarantee the inclusion of critical elements, such as tumor measurements and response assessments, according to Response Evaluation Criteria in Solid Tumors (RECIST) guidelines.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    One study reported a 30% reduction in initial report generation time with AI report drafting, though comprehensive evaluation including verification and correction time remains limited.

    “One study reported a 30% reduction in initial report generation time, though comprehensive evaluation including verification and correction time by radiologists remains limited [56].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] Wornow, 2025, Zero-Shot Clinical Trial Patient Matching with LLMs, NEJM AI source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationreasoning

    AI moves beyond optimizing existing tasks and supports identification of previously unrecognized biological relationships and design of new diagnostic and treatment approaches.

    “By analyzing complex, high-dimensional data, AI moves beyond optimizing existing tasks and supports the identification of previously unrecognized biological relationships and the design of new diagnostic and treatment approaches.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    TOAD can prioritize ancillary testing and assist in the diagnostic workup by providing slide-level attention maps and calibrated probabilities.

    “By providing slide-level attention maps and calibrated probabilities, TOAD can prioritize ancillary testing and assist in the diagnostic workup.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    PanDerm, trained on >2 million dermoscopic and clinical images, improved physicians' diagnostic accuracy for skin cancer by >10%, with the greatest benefits for non-dermatologists.

    “a foundation model such as PanDerm, trained on >2 million dermoscopic and clinical images, improved physicians’ diagnostic accuracy for skin cancer by >10%, with the greatest benefits observed for non-dermatologists [26].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] Yan, 2025, A Multimodal Vision Foundation Model for Clinical Dermatology, Nat. Med. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Woollie, an oncology-specific LLM, achieved an AUROC of 0.97 for progression prediction on internal validation and 0.88 on external validation at UCSF.

    “trained on real-world data from Memorial Sloan Kettering Cancer Center across lung, breast, prostate, pancreatic, and colorectal cancers, it achieved an area under the receiver operating characteristic curve (AUROC) of 0.97 for progression prediction, with external validation at University of California, San Francisco (UCSF) showing an AUROC of 0.88 [50].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [50] Zhu, 2025, Large Language Model Trained on Clinical Oncology Data Predicts Cancer Progression, npj Digit. Med. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Virtual biopsies may reduce the need for an initial biopsy; histology-derived molecular surrogates can reduce add-on assays and cost while still relying on tissue obtained for standard care.

    “The former may reduce the need for an initial biopsy; the latter can reduce add-on assays and cost while still relying on tissue obtained for standard care.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    MUSK, a vision-language system, pretrains on large, largely unpaired corpora of pathology image patches and clinical text, then aligns image-text pairs for downstream tasks.

    “The Multimodal transformer with Unified maSKed modeling (MUSK) from Stanford exemplifies this approach as a vision-language system that pretrains on large, largely unpaired corpora of pathology image patches and clinical text, then aligns image–text pairs for downstream tasks [32].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [32] Xiang, 2025, A Vision-Language Foundation Model for Precision Oncology, Nature source

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    Both virtual biopsies and histology-derived surrogates require prospective, multi-site validation before routine adoption.

    “Both require prospective, multi-site validation before routine adoption.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Pathology AI performance varied by subspecialty: gastrointestinal (93%/94%), urologic (95%/96%), and breast (83%/88%).

    “Performance varied by subspecialty: gastrointestinal (93%/94%), urologic (95%/96%), and breast (83%/88%) [35].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [35] McGenity, 2024, Artificial Intelligence in Digital Pathology: A Systematic Review and Meta-Analysis of Diagnostic Test Accuracy, npj Digit. Med. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    LLM-based chat systems can generate plausible-sounding but factually incorrect information with high confidence, potentially leading to dangerous clinical decisions.

    “LLM-based chat systems can generate plausible-sounding but factually incorrect information with high confidence, potentially leading to dangerous clinical decisions.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    AlphaFold predicts a protein's most stable shape but does not model how the protein moves.

    “It predicts a protein’s most stable shape; however, it does not model how the protein moves.”
    Quote in the paperSubstantial claim0 of 4 sources checked
    • cites [105] Nussinov, 2023, AlphaFold, Allosteric, and Orthosteric Drug Discovery: Ways Forward, Drug Discov. Today source

      Not checked in the free run.

    • cites [106] Terwilliger, 2024, AlphaFold Predictions Are Valuable Hypotheses and Accelerate but Do Not Replace Experimental Structure Determination, Nat. Methods source

      Not checked in the free run.

    • cites [107] Lekadir, 2025, FUTURE-AI: International Consensus Guideline for Trustworthy and Deployable Artificial Intelligence in Healthcare, BMJ source

      Not checked in the free run.

    • cites [108] Daneshjou, 2022, Disparities in Dermatology AI Performance on a Diverse, Curated Clinical Image Set, Sci. Adv. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    SCORPIO was developed on an institutional cohort (n≈1600) and validated across internal sets (n≈2500), multicenter phase-3 trials (n≈4400), and an external real-world cohort (n≈1100), covering nearly 10,000 ICI-treated patients across more than 20 cancer types.

    “Developed on an institutional cohort (n ≈ 1600) and validated across internal sets (n ≈ 2500), multicenter phase-3 trials (n ≈ 4400), and an external real-world cohort (n ≈ 1100), the SCORPIO model covers nearly 10,000 immune checkpoint inhibitors (ICI)-treated patients across more than 20 types of cancers.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [83] Aleksakhina, S.N., and Imyanitov, E.N. (2021). Cancer Therapy Guided by Mutation Tests: Current Status and Perspectives. Int. J. Mol. Sci., 22. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    For WSB1, investigators built an AlphaFold2 model, refined it with molecular simulation, and conducted virtual screening across millions of compounds.

    “Investigators built an AlphaFold2 model of WSB1, refined it with molecular simulation, and conducted virtual screening across millions of compounds.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [104] Agarwal, 2024, The Power and Pitfalls of AlphaFold2 for Structure Prediction beyond Rigid Globular Proteins, Nat. Chem. Biol. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    This hybrid approach enables discovery of novel patterns while respecting established mechanistic understanding.

    “This hybrid approach enables discovery of novel patterns while respecting established mechanistic understanding.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Approximately 20 WSB1 candidates were selected for synthesis and testing, yielding four lead molecules.

    “Approximately 20 candidates were selected for synthesis and testing, yielding four lead molecules.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [104] Agarwal, 2024, The Power and Pitfalls of AlphaFold2 for Structure Prediction beyond Rigid Globular Proteins, Nat. Chem. Biol. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    AlphaFold removes a frequent bottleneck in structure-based design and makes structural hypotheses more accessible.

    “By accurately predicting three-dimensional structures from sequences at the proteome scale, AlphaFold removes a frequent bottleneck in structure-based design and makes structural hypotheses more accessible.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [101] Le, M.H.N., Nguyen, P.K., Nguyen, T.P.T., Nguyen, H.Q., Tam, D.N.H., Huynh, H.H., Huynh, P.K., and Le, N.Q.K. (2025). An In-Depth Review of AI-Powered Advancements in Cancer Drug Discovery. Biochim. Biophys. Acta Mol. Basis Dis., 1871. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Seven of the approximately 9000 candidate molecules were synthesized and tested; one showed promising activity in liver cancer cells and advanced as a lead, all within roughly 30 days from design initiation.

    “Seven of these molecules were synthesized and tested; one showed promising activity in liver cancer cells and advanced as a lead, all within roughly 30 days from design initiation.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [102] Weng, 2022, Identification of Potential WSB1 Inhibitors by AlphaFold Modeling, Virtual Screening, and Molecular Dynamics Simulation Studies, Evid. Based Complement. Altern. Med. source

      Not checked in the free run.

    • cites [103] Schauperl, 2022, AI-Based Protein Structure Prediction in Drug Discovery: Impacts and Challenges, J. Chem. Inf. Model. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    By monitoring multiple physiological scales simultaneously, deep phenotyping can detect early transitions from wellness to disease and inform personalized treatment decisions based on the individual's complete biological context rather than tumor characteristics alone.

    “By monitoring multiple physiological scales simultaneously, deep phenotyping can detect early transitions from wellness to disease and inform personalized treatment decisions based on the individual’s complete biological context rather than tumor characteristics alone.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationcitation

    This holistic approach represents a paradigm shift toward P4 medicine, which could transform cancer care by enabling earlier intervention and more precise therapeutic matching.

    “This holistic approach represents a paradigm shift toward predictive, preventive, personalized, and participatory (P4) medicine, which could transform cancer care by enabling earlier intervention and more precise therapeutic matching [99,100].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [99] Jumper, 2021, Highly Accurate Protein Structure Prediction with AlphaFold, Nature source

      Not checked in the free run.

    • cites [100] Ren, 2023, AlphaFold Accelerates Artificial Intelligence Powered Drug Discovery: Efficient Discovery of a Novel CDK20 Small Molecule Inhibitor, Chem. Sci. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    By quantifying disease heterogeneity and predicting individual trajectories, AI-enhanced deep phenotyping moves us closer to the dual goals of precision oncology: optimal treatment selection for existing cancers and proactive identification of at-risk individuals for prevention.

    “By quantifying disease heterogeneity and predicting individual trajectories, AI-enhanced deep phenotyping moves us closer to the dual goals of precision oncology: optimal treatment selection for existing cancers and proactive identification of at-risk individuals for prevention.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    WSB1 is a protein linked to metastasis through hypoxia-response pathways in lung and pancreatic cancers and historically underexplored as a drug target.

    “A second example concerns WSB1, a protein linked to metastasis through hypoxia-response pathways in lung and pancreatic cancers but historically underexplored as a drug target [104].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [104] Agarwal, 2024, The Power and Pitfalls of AlphaFold2 for Structure Prediction beyond Rigid Globular Proteins, Nat. Chem. Biol. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Drug binding often depends on protein motion, such as small shifts when a compound binds, long-range effects within the protein, hidden pockets that open and close, or changes when proteins form complexes.

    “Drug binding often depends on motion, such as small shifts when a compound binds, long-range effects within the protein, hidden pockets that open and close, or changes when proteins form complexes.”
    Quote in the paperSubstantial claim0 of 4 sources checked
    • cites [105] Nussinov, 2023, AlphaFold, Allosteric, and Orthosteric Drug Discovery: Ways Forward, Drug Discov. Today source

      Not checked in the free run.

    • cites [106] Terwilliger, 2024, AlphaFold Predictions Are Valuable Hypotheses and Accelerate but Do Not Replace Experimental Structure Determination, Nat. Methods source

      Not checked in the free run.

    • cites [107] Lekadir, 2025, FUTURE-AI: International Consensus Guideline for Trustworthy and Deployable Artificial Intelligence in Healthcare, BMJ source

      Not checked in the free run.

    • cites [108] Daneshjou, 2022, Disparities in Dermatology AI Performance on a Diverse, Curated Clinical Image Set, Sci. Adv. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Cluster 1 showed very low recurrence, whereas cluster 2 had higher recurrence and mortality, indicating a previously unrecognized subtype linked to prognosis based solely on preoperative data.

    “Cluster 1 showed very low recurrence, whereas cluster 2 had higher recurrence and mortality, indicating a previously unrecognized subtype linked to prognosis based solely on preoperative data.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [87] Fujita, 2022, Proteo-Genomic Characterization of Virus-Associated Liver Cancers Reveals Potential Subtypes and Therapeutic Targets, Nat. Commun. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    The FUTURE-AI framework was developed by 117 experts from 50 countries and offers six practical principles: fairness, universality, traceability, usability, robustness, and explainability.

    “The FUTURE-AI framework, developed by 117 experts from 50 countries, offers six practical principles to guide this work: fairness, universality, traceability, usability, robustness, and explainability [109] (Figure 5).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [109] Adamson, 2018, Machine Learning and Health Care Disparities in Dermatology, JAMA Dermatol. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Regulatory frameworks for AI in oncology remain evolving, with significant gaps in liability assignment when AI-assisted decisions lead to adverse outcomes.

    “Beyond technical challenges, regulatory frameworks for AI in oncology remain evolving, with significant gaps in liability assignment when AI-assisted decisions lead to adverse outcomes [124].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [124] Tellez, 2019, Quantifying the Effects of Data Augmentation and Stain Color Normalization in Convolutional Neural Networks for Computational Pathology, Med. Image Anal. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Traceability systems generate regulatory-ready documentation, streamlining the path from research to clinical deployment.

    “These systems generate regulatory-ready documentation, streamlining the path from research to clinical deployment.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Data poisoning attacks could corrupt training data and adversarial attacks are designed to cause misclassification at inference time.

    “Furthermore, models must be resilient against deliberate threats, including data poisoning attacks that could corrupt the training data or adversarial attacks designed to cause misclassification at inference time [152].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [152] Salahuddin, Z., Woodruff, H.C., Chatterjee, A., and Lambin, P. (2022). Transparency of Deep Neural Networks for Medical Image Analysis: A Review of Interpretability Methods. Comput. Biol. Med., 140. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    SHAP values can rank the importance of different clinical factors and show whether a prognosis was driven primarily by tumor size, patient age, or biomarker status.

    “For tabular data, SHAP values can rank the importance of different clinical factors and show whether a prognosis was driven primarily by tumor size, patient age, or biomarker status [162,163,164,165].”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [162] Rudin, 2019, Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead, Nat. Mach. Intell. source

      Not checked in the free run.

    • cites [163] Amann, J., Blasimme, A., Vayena, E., Frey, D., and Madai, V.I. (2020). Precise4Q consortium Explainability for Artificial Intelligence in Healthcare: A Multidisciplinary Perspective. BMC Med. Inform. Decis. Mak., 20. source

      Not checked in the free run.

    • cites [164] Spaanderman, D.J., Marzetti, M., Wan, X., Scarsbrook, A.F., Robinson, P., Oei, E.H.G., Visser, J.J., Hemke, R., van Langevelde, K., and Hanff, D.F. (2025). AI in Radiological Imaging of Soft-Tissue and Bone Tumours: A Systematic Review Evaluating against CLAIM and FUTURE-AI Guidelines. eBioMedicine, 114. source

      Not checked in the free run.

    • cites [165] Fiske, 2025, Climate Change and Health: The next Challenge of Ethical AI, Lancet Glob. Health source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Continuous monitoring systems detect when models encounter data outside their training distribution or when performance begins to degrade.

    “Continuous monitoring systems detect when models encounter data outside their training distribution or when performance begins to degrade [147,148,149,150,151].”
    Quote in the paperNot a substantial claim0 of 5 sources checked
    • cites [147] Guan, H., Bates, D., and Zhou, L. (2020). Keeping Medical AI Healthy: A Review of Detection and Correction Methods for System Degradation. arXiv.

      Not checked in the free run.

    • cites [148] Qazi, M.A., Hashmi, A.U.R., Sanjeev, S., Almakky, I., Saeed, N., Gonzalez, C., and Yaqub, M. (2025). Continual Learning in Medical Imaging: A Survey and Practical Analysis. arXiv.

      Not checked in the free run.

    • cites [149] Perkonigg, 2021, Dynamic Memory to Alleviate Catastrophic Forgetting in Continual Learning with Medical Imaging, Nat. Commun. source

      Not checked in the free run.

    • cites [150] Qayyum, 2021, Secure and Robust Machine Learning for Healthcare: A Survey, IEEE Rev. Biomed. Eng. source

      Not checked in the free run.

    • cites [151] Adadi, 2018, Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI), IEEE Access source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given

    AI can automate diagnostic tasks, streamline workflows, and enable novel insights through computational biomarkers and drug discovery.

    “While AI can automate diagnostic tasks, streamline workflows, and enable novel insights through computational biomarkers and drug discovery, its successful clinical implementation requires more than technical accuracy.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    Comprehensive traceability enables rapid identification of problems, facilitates regulatory approval, and builds institutional trust.

    “Comprehensive traceability serves multiple purposes: enabling rapid identification of problems, facilitating regulatory approval, and building institutional trust.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Training a single large language model generates emissions equivalent to approximately 300,000 kg of CO2, which is roughly five times the lifetime emissions of an average automobile.

    “Training a single large language model generates emissions equivalent to approximately 300,000 kg of CO2, which is roughly five times the lifetime emissions of an average automobile [165,166].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [165] Fiske, 2025, Climate Change and Health: The next Challenge of Ethical AI, Lancet Glob. Health source

      Not checked in the free run.

    • cites [166] Hao, K. (2025, October 01). Training a Single AI Model Can Emit as Much Carbon as Five Cars in Their Lifetimes. Available online: https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationnone given

    Successful clinical implementation of AI in oncology requires more than technical accuracy.

    “While AI can automate diagnostic tasks, streamline workflows, and enable novel insights through computational biomarkers and drug discovery, its successful clinical implementation requires more than technical accuracy.”
    Quote in the paperNot a substantial claim
51 of 51

  • The paper relies on this without support.

    The FUTURE-AI framework provides practical guidelines to ensure AI tools are fair, reliable, and understandable in clinical practice.

    “The FUTURE-AI framework provides practical guidelines to ensure these AI tools are fair, reliable, and understandable when used in clinical practice.”

  • The paper relies on this without support.

    AI augments rather than replaces physician expertise and clinical judgment.

    “This paper highlights how AI augments rather than replaces physician expertise”
    “which augments rather than replaces clinical judgment”

  • The paper relies on this without support.

    Drug discovery has long been slow, costly, and associated with high attrition rates.

    “Drug discovery has long been slow, costly, and associated with high attrition rates.”

  • The paper relies on this without support.

    AI use in clinical oncology will continue to grow.

    “AI helps clinical oncology with both productivity and creativity, and its use will continue to grow.”
4 of 4

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • FUTURE-AI review cites a reference about carbon emissions

    Minor

    The cited reference [166] is about the carbon emissions of training an AI model, not a systematic review of FUTURE-AI scores. The reference list identifies the soft-tissue and bone-tumor review as [164].

1 of 1

Cites

Flagged first, then by citations

1 of the 159 references couldn’t be matched to a known paper.

Research works printed but not linked: 7

  1. Werbos, P.J. (1990, January 4–6). Backpropagation through time: What it does and how to do it. Proceedings of the IEEE, Skokie, IL, USA.
  2. Sapkota, R., Roumeliotis, K.I., and Karkee, M. (2025). Vibe Coding vs. Agentic Coding: Fundamentals and Practical Implications of Agentic AI. arXiv.
  3. Yurkovich, 2020, A Systems Approach to Clinical Oncology Uses Deep Phenotyping to Deliver Personalized Care. Nature Reviews, Clin. Oncol., 17, 183
  4. Yurkovich, 2024, The Transition from Genomics to Phenomics in Personalized Population Health. Nature Reviews, Genetics, 25, 286
  5. Schelter, L., and Klein, T. (2025, October 20). Automatically Tracking Metadata and Provenance of Machine Learning Experiments. In Machine Learning Systems Workshop at NIPS. Available online: https://neurips.cc/virtual/2017/workshop/8774.
  6. Guan, H., Bates, D., and Zhou, L. (2020). Keeping Medical AI Healthy: A Review of Detection and Correction Methods for System Degradation. arXiv.
  7. Qazi, M.A., Hashmi, A.U.R., Sanjeev, S., Almakky, I., Saeed, N., Gonzalez, C., and Yaqub, M. (2025). Continual Learning in Medical Imaging: A Survey and Practical Analysis. arXiv.

Cited by

Sorted by citations

Authors
Kuno, M., Osumi, H., Udagawa, S., Yoshikawa, K., Ooki, A., Shinozaki, E., Ishikawa, T., Oba, J., Yamaguchi, K., & Sakurada, K.
Affiliations
  • Keio University (JP)
  • The Cancer Institute Hospital (JP)
Corresponding author
  • Osumi, Hiroki
  • Sakurada, Kazuhiro
ORCID
Venue
Current Oncology
Publisher
MDPI AG
Published
2025-10-22
Volume
32
Issue
11
Pages
588
ISSN
  • 1198-0052 (ISSN-L)
  • 1718-7729
DOI
10.3390/curroncol32110588
PMC
PMC12651923
OpenAlex
W4415421320
Cited by
7
References
159
Citations by year
2026: 7
FWCI
0.41
Type
journal article
Language
en
Open access
Licence
Locations
  • Current Oncology (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Artificial Intelligence › AI in cancer detection
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
Keywords
artificial intelligence in oncology, clinical productivity, AI-assisted diagnosis, clinical documentation automation, deep learning, transformer foundation models, multimodal data integration, anticancer drug development, clinical translation of AI, responsible AI deployment, patient-centered care
MeSH terms
Artificial Intelligence, Humans, Medical Oncology, Neoplasms

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-07.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-07.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review