AI and Machine Learning in Oncology

Source-checked note: this page is about research, software, and regulated clinical decision support; regulatory links were checked on 2026-05-22. It does not endorse using unvalidated models for diagnosis or treatment selection.

TL;DR

AI is genuinely useful in oncology, especially in imaging workflows, pathology quantification, trial matching, genomics support, operations, and research discovery. The conservative rule is simple: a model is not clinically trustworthy because it has a high AUC; it becomes useful only when validated for the population, workflow, endpoint, and regulatory context where it will be used.

Where AI Is Most Mature

Area	Maturity
Radiology and screening support	Many AI/ML-enabled medical devices appear on FDA device lists; value depends on task and workflow
Digital pathology	Increasing deployment for quantification, triage, and assistive review; clinical claims are tool-specific
Genomic interpretation	Common in variant prioritization and annotation pipelines, usually as decision support
Trial matching	Operationally useful when connected to curated eligibility criteria and EHR data
Drug discovery	Strong impact on early discovery; patient benefit still requires pharmacology and trials
Prognosis or treatment-response prediction	High publication volume, but fewer externally validated, prospectively deployed tools

Evaluation Standard

Before publishing a clinical-sounding claim, require:

1. Patient-level train/validation/test splits.
2. External validation by site, time, or geography.
3. Calibration, not only discrimination.
4. Subgroup performance across sex, age, ancestry, scanner, hospital, and disease subtype when relevant.
5. Clear decision threshold and intended user.
6. Prospective evidence when the tool changes patient management.

Regulatory Context

• United States: FDA tracks AI-enabled medical devices and has final guidance for Predetermined Change Control Plans (PCCPs) in AI-enabled device software functions.
• European Union: the AI Act entered the regulatory landscape in 2024 and treats many medical AI systems as high-risk.
• Brazil: software as a medical device falls under ANVISA medical-device regulation, including RDC 657/2022 and RDC 751/2022 contexts; LGPD remains central for patient data.

Common Failure Modes

• Data leakage: slices, patches, or visits from the same patient split across train and test.
• Site leakage: scanner or hospital signature becomes the model.
• Label bias: historical care inequity becomes ground truth.
• Missing calibration: a risk score looks precise but does not match observed probability.
• No deployment monitoring: drift appears after a new scanner, protocol, population, or treatment pathway.

What Technologists Can Build

• Reproducible data pipelines with lineage and versioned reference data.
• Evaluation harnesses that report calibration and subgroup performance by default.
• Human-in-the-loop interfaces that show uncertainty and escalation paths.
• Model monitoring for drift, alerts, rollback, and audit logs.
• Trial-matching tools that cite exact eligibility criteria and registry records.

See Also

• ML pitfalls in oncology
• Multi-omics integration
• Network biology
• Precision medicine
• Data governance and LGPD

References

1. U.S. Food and Drug Administration. Artificial Intelligence-Enabled Medical Devices. https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices
2. U.S. Food and Drug Administration. Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions. Final guidance, August 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence
3. European Union. Regulation (EU) 2024/1689: Artificial Intelligence Act. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
4. ANVISA. Software como Dispositivo Médico: perguntas e respostas RDC 657/2022. https://www.gov.br/anvisa/pt-br/assuntos/noticias-anvisa/2022/software-como-dispositivo-medico-perguntas-e-respostas/perguntas-respostas-rdc-657-de-2022-v1-01-09-2022.pdf/view
5. Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. https://doi.org/10.1136/bmj-2023-078378