Breast Cancer and Artificial Intelligence: From Risk Prediction to Treatment Simulation with Digital Twins
In this article
The AI Revolution in Breast Cancer Management: From Risk Prediction to Treatment Simulation with a “Digital Twin”
The Journey from Statistics to Prediction
The central paradigm of modern oncology is moving away from a “one-size-fits-all” approach and tailoring treatment to the unique biological characteristics of the individual and the tumor. Breast cancer is among the clearest examples of this heterogeneity; beyond major subtypes such as luminal, HER2-positive and triple-negative (TNBC), each tumor has its own genomic, transcriptomic and microenvironmental signature. Clinical management involves interconnected, multivariable decision networks, including risk assessment, screening strategies, neoadjuvant or adjuvant treatment decisions, surgical timing and recurrence monitoring.
For years, we relied on the limits of human cognition and simple statistical models to manage this complexity. Biological systems, however, are not linear; interactions among genes, proteins and the tissue microenvironment are chaotic and multidimensional. Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL) algorithms, has therefore become an indispensable tool for managing this complexity.
The history of breast cancer management can be divided into three main eras:
- 1. The Statistical Era (1989-2000): The period of population-based statistics, such as the Gail Model.
- 2. The Genomic Era (2000-2015): The period when genetic risk factors entered the equation, as with Tyrer-Cuzick.
- 3. The Artificial Intelligence (AI) Era (2016-Present): The period when deep learning algorithms “predict the future” by processing pixels and biological data imperceptible to the human eye.
This article examines in depth how AI is transforming breast cancer management, from the Mirai algorithm to virtual surgical planning, using the latest technologies, evidence-based data and regulatory (FDA, ESMO) frameworks.
1. A Revolution in Risk Prediction: From Questionnaires to Pixels
Conventional risk models (Gail, Tyrer-Cuzick) rely on “static,” questionnaire-based information such as age, family history and number of births. Their main limitation is their inability to fully use biological clues within breast tissue itself (density, microcalcifications, etc.). AI has transformed mammography from a “photograph” showing current cancer into a “crystal ball” revealing the future.
Mirai vs. Conventional Models: Comparative Performance (AUC)
Journal of Clinical Oncology and Radiology According to validation studies published in these journals:
Among patients who would develop cancer in the future, 31% were correctly identified as “high risk.”
Among patients who would develop cancer in the future, only 18% were identified.
The most basic statistical model is inadequate for predicting modern risks.
Why the Difference? Mirai can analyze tissue microstructures, asymmetries and very subtle changes between previous and current mammograms (temporal data) that the human eye cannot detect.
Technology with FDA “Breakthrough” Designation: Prognosia Breast
Developed by Washington University, Prognosia Breastreceived FDA “Breakthrough Device” designation. This technology measures not only the amount of breast density but also its distribution and microstructural irregularities (texture analysis).
Even before a tumor forms, this analysis can detect a microenvironment conducive to tumor development (pre-tumor environment) in breast tissue. Prognosia Breast has an AUC score ranging from 0.67 to 0.74. Validation in large datasets including women of different ethnic backgrounds (Black, Asian) also reduces the risk of “data bias” in AI.
The Shift to Personalized Screening: The End of “One-Size-Fits-All Screening”
AI-based risk scoring enables optimal resource use. Instead of standard annual mammography for every woman over 40, dynamic risk-based protocols can now be used. According to University of Cambridge and RSNA 2025 data, AI can predict interval cancer risk by analyzing hidden patterns in mammograms reported as negative.
- High-Risk Patient (High AI Score): Screening becomes more frequent (e.g., every 6 months). Breast MRI or contrast-enhanced mammography is added. The aim is to detect interval cancers.
- Low-Risk Patient (Very Low AI Score): Screening intervals can safely be extended (e.g., every 2 years), reducing unnecessary radiation exposure, anxiety and the financial burden on the healthcare system.
2. The Future of Treatment: Digital Twins and Virtual Trials
Imagine a future in which patients no longer need to be “guinea pigs” to select the right drug. AI makes this possible by simulating treatments on a “Digital Twin” created from the patient's biological data.
FarrSight-Twin: From Astrophysics to Oncology
Adapting complex astrophysical algorithms used in the discovery of black holes to cancer biology, this platform processes data from thousands of patients to create a personalized “virtual avatar” for a new patient.
Clinical Application: In challenging cases with limited treatment options, such as triple-negative breast cancer (TNBC), it tests which drug combination (chemotherapy + immunotherapy, etc.) will produce the best outcome in silico (in a computer environment) . It also enables “Virtual Clinical Trials” to reduce drug development costs.
SimBioSys TumorSight: Virtual Surgical Planning
This FDA-authorized (510(k) Cleared) platform uses the patient's dynamic contrast-enhanced breast MRI (DCE-MRI) images to generate a 3-dimensional biophysical tumor model. The model simulates not just shape but also tumor growth kinetics.
What It Means for the Surgeon: Before operating, the surgeon can see the answer to “Can I achieve clear surgical margins with breast-conserving surgery?” in this simulation. The goal is to minimize the risk of repeat surgery (re-excision) and breast loss.
3. Predicting Treatment Response and Resistance: Decoding Biological Signals
AI explains not only “what will happen” (the outcome) but also “why it will happen” (the mechanism). Two key developments decoding the tumor ecosystem are:
- Multi-Omics Integration (Nature, 2022): Predicting treatment response using clinical data alone is difficult. In the study by Sammut and colleagues published in Nature, the AI model combined age, tumor DNA (mutation burden, copy number variations), RNA (gene expression) and digital pathology images (immune cell infiltration - TILs).
Result: This integration predicted whether the patient would achieve a complete response to chemotherapy (neoadjuvant chemotherapy - NAC) with 0.87 AUC , a very high level of accuracy. The model also identified “immune escape” mechanisms in treatment-resistant tumors (such as loss of HLA class I alleles), revealing new drug targets. - The PERCI Index (Endocrine Resistance): Some patients with hormone receptor-positive (HR+) breast cancer develop resistance to hormonal treatment over time. The PERCI (Predictive Endocrine Resistance Index) score uses machine learning to analyze data such as TP53 mutation status, DNA methylation patterns and age.
Clinical Benefit: It answers the physician's question: “Is endocrine therapy alone sufficient for this patient, or should a targeted drug (CDK4/6 inhibitor) be added from the outset?”
4. Implications for Everyday Practice: AI in the Clinic
AI now supports physicians in everyday clinical workflows as well as in laboratories.
| Technology / Software | Application | Clinical Benefit and Function |
|---|---|---|
| Koios DS (FDA 510(k) Cleared) | Breast Ultrasound | When a radiologist marks a suspicious lesion, it analyzes it within seconds, provides a BI-RADS-compatible risk score and recommends “biopsy required” or “follow-up sufficient.” It has been shown to increase radiologists' diagnostic accuracy (AUC) from 0.83 to 0.87. |
| GUIDE-G / RAG Technology | Guideline Integration | It searches complex treatment guidelines spanning hundreds of pages, such as NCCN and ASCO, to instantly present the most current, evidence-based treatment option for a patient. The RAG-based system developed by the National University of Singapore minimizes the risk of AI “hallucination” (incorrect information). |
| SurvGPT | Survivorship Care | Automatically prepares a “Survivorship Care Plan” (SCP) for patients who have completed treatment. It searches Electronic Health Records (EHR), summarizes chemotherapy doses, radiation fields and adverse effects, and creates a personalized follow-up schedule. |
| TabNet | Recurrence Prediction | Predicts recurrence risk “with explanations” (explainable AI) by analyzing not just the tumor but also contralateral breast symmetry and surrounding (peritumoral) tissue. |
5. Regulatory Framework: Safe AI Use
Regulatory agencies (FDA) and professional organizations (ESMO) establish standards for safely integrating AI into clinical care.
- FDA (US Food and Drug Administration): Evaluates AI algorithms as “Software as a Medical Device” (SaMD). For self-learning algorithms, it requires a “Predetermined Change Control Plan” (PCCP) that monitors changes in performance over time.
- ESMO (European Society for Medical Oncology): Has published two key guidance documents:
- ELCAP: Sets boundaries for using Large Language Models, requiring source verification and the “human-in-the-loop” principle.
- EBAI: Defines the analytical and clinical validity criteria required for AI-based biomarkers to enter clinical practice.
Scientific References and Further Reading
- Yala A, Mikhael PG, Strand F, et al. Toward robust mammography-based models for breast cancer risk (Mirai). Science Translational Medicine. 2021;13(578):eaba4373.
- Sammut SJ, Crispin-Ortuzar M, Chin SF, et al. Multi-omic machine learning predictor of breast cancer therapy response. Nature. 2022;601(7894):623-629.
- Lehman CD, Yala A, Schuster T, et al. Mammographic Breast Density Assessment Using Deep Learning: Clinical Implementation. Radiology. 2019;290(1):52-58.
- Harbeck N, et al. A predictive endocrine resistance index accurately stratifies luminal breast cancer treatment responders and nonresponders (PERCI). J Clin Invest. 2024.
- SimBioSys. FDA Clears TumorSight Viz for Breast Cancer Surgical Planning. Press Release, 2023.
- ESMO Real World Data and Digital Health Working Group. Guidance on the use of Large Language Models in Oncology. Annals of Oncology. 2024.
*This article draws on Prof. Dr. Mustafa Özdoğan's presentation titled “Meme Kanseri Yönetiminde Yapay Zeka: Risk Tahmininden Tedavi Simülasyonuna” and the current literature.
Was this article helpful?
Your feedback helps us improve our content.
Reader comments
Share your comment
Your name and comment are published only after administrator approval. Your email address is never shown to visitors.
Be the first to comment.