Loading
Radiomics and Machine Learning in Medical Imaging / Ai in Cancer Detection / Breast Cancer Treatment Studies · Journal article
Cancers · September 11, 2026
Early or partial results. Treat as a signal, not a conclusion.
This systematic review of 102 AI studies for breast cancer neoadjuvant response prediction found that the evidence base remains immature: no models have reached prospective workflow integration or interventional validation, most carry high bias risk, and external validation is reported in only half. The review concludes that current AI evidence does not yet support treatment tailoring (drug omission, switching, escalation, or de-escalation) based on model prediction.
Systematic review of AI prediction studies with structured critical appraisal. Full-text studies of artificial intelligence models predicting response to neoadjuvant therapy in breast cancer (any subtype), published 2020–2026.. Intervention: Artificial intelligence models for predicting pathologic complete response (pCR) or treatment response under fixed regimens. n = 102.
43 of 102 studies reached internal validation only (L1) 53 of 102 studies achieved temporal or geographic external validation (L2) 6 of 102 studies reached prospective observational validation (L3)
Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.
Clinicians should not use current AI models to guide individual treatment tailoring (drug omission, switching, escalation, de-escalation) in neoadjuvant breast cancer. The evidence is too early-stage, with high bias risk and no prospective or interventional validation; treatment decisions should remain guided by established protocols and clinical judgement.
A systematic review of 102 AI prediction studies showing that most lack external validation, prospective evidence, or workflow integration; none demonstrate interventional benefit for treatment tailoring in breast cancer neoadjuvant therapy.
As stated by the source record.
Quoted from the source exactly as published.
Clinicians should not use current AI models to guide individual treatment tailoring (drug omission, switching, escalation, de-escalation) in neoadjuvant breast cancer. The evidence is too early-stage, with high bias risk and no prospective or interventional validation; treatment decisions should remain guided by established protocols and clinical judgement.
Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.
Neoadjuvant therapy for breast cancer is planned by subtype: pathologic complete response (pCR) differs in frequency, meaning, and surrogate validity across HR+/HER2−, HER2+, and triple-negative breast cancer (TNBC). This review examines whether current evidence supports using artificial intelligence (AI) to move beyond predicting response under a fixed regimen to guiding systemic treatment tailoring. Here, tailoring means model-guided drug omission, switching, escalation, or de-escalation for an individual patient. We appraised 102 full-text studies of AI-based response prediction (2020–2026), organized by subtype and clinical decision, and graded each on an author-defined five-level clinical-readiness ladder (L1–L5) measuring validation and translational maturity rather than accuracy. Risk of bias was assessed with PROBAST and reporting against TRIPOD+AI. Readiness clustered low: 43 studies reached internal validation only (L1), 53 temporal or geographic external validation (L2), and 6 prospective observational validation (L3); none reached workflow integration (L4) or interventional evidence (L5). Most carried high overall risk of bias (89/102); external validation appeared in 52 (51%), fully reported decision-curve analysis in 44 (43%), and calibration in 19 (19%). Strategy comparison and individualized-treatment-effect analyses were essentially absent. Subtype-specific AI currently predicts response under fixed regimens with moderate-to-good discrimination; the evidence does not yet support changing an individual patient’s systemic treatment.
Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.