ML Predicts 3-Year Survival in Locally Advanced Breast Cancer

Locally advanced breast cancer (LABC) affects over 200,000 people globally each year, with 3-year survival outcomes varying wildly even among patients receiving identical neoadjuvant chemotherapy (NAC) regimens.

Current survival prediction tools rely heavily on invasive tissue biopsies and limited clinical markers, leaving clinicians with incomplete information to guide treatment plans. A new breakthrough uses machine learning (ML) and quantitative ultrasound (QUS) imaging to deliver accurate, non-invasive 3-year survival predictions for LABC patients.

What Is Locally Advanced Breast Cancer?

LABC refers to breast cancer that has spread to nearby lymph nodes or chest tissue but has not metastasized to distant organs. It accounts for roughly 30% of all new breast cancer diagnoses in low- and middle-income countries.

Neoadjuvant chemotherapy (NAC) is the standard first-line treatment for LABC: it shrinks tumors before surgery, making them easier to remove, and helps eliminate microscopic cancer cells spreading through the body.

3-year survival rates for LABC patients range from 50% to 80% depending on how well their tumors respond to NAC. Accurately predicting a patient’s survival early in treatment can help clinicians adjust care plans to improve outcomes.

Quantitative Ultrasound Imaging: A Non-Invasive Game Changer

Traditional breast ultrasound is qualitative: radiologists visually assess images to look for suspicious masses, a process that is subjective and can miss subtle tumor characteristics.

Quantitative ultrasound (QUS) extracts objective, numeric features from ultrasound wave data. It measures factors like tissue stiffness, echogenicity (how sound waves bounce off tissue), and texture patterns that are invisible to the naked eye.

QUS has major advantages over other imaging modalities: it uses no ionizing radiation, costs a fraction of MRI scans, and can be performed during routine clinic visits with no downtime for patients.

How Machine Learning Boosts Survival Prediction Accuracy

Historically, clinicians have used basic clinical factors—age, tumor size, lymph node involvement—to estimate survival risk. These methods have limited accuracy, with area under the curve (AUC) scores often below 0.7.

Machine learning models can analyze thousands of QUS features simultaneously, identifying complex patterns that human researchers cannot detect. In recent validation studies, ML models trained on QUS data achieved AUC scores above 0.85 for predicting 3-year survival in LABC patients receiving NAC.

Common ML algorithms used for this application include random forest classifiers, support vector machines, and deep neural networks, all of which were found to outperform traditional prediction methods in head-to-head testing.

Top QUS Features Linked to Survival Risk

• Tumor texture heterogeneity: More irregular, varied tissue patterns correlate with higher risk of poor survival outcomes.
• Tissue stiffness metrics: Harder tumor tissue, measured via QUS elastography, is associated with more aggressive cancer behavior.
• Vascularity features: Abnormal blood flow patterns around the tumor indicate faster growth and higher metastasis risk.
• Lymph node characteristics: QUS features of nearby affected lymph nodes provide critical data on how far cancer has spread locally.

Real-World Benefits for Patients and Clinicians

This ML-QUS approach delivers tangible value across the care journey:

• Personalized treatment adjustments: Patients predicted to have low 3-year survival can be switched to more aggressive NAC regimens, enrolled in clinical trials, or given targeted therapies earlier.
• Reduced overtreatment: Patients with high predicted survival can avoid unnecessary aggressive treatments that cause severe side effects.
• Non-invasive monitoring: Repeated QUS scans can track how a tumor responds to NAC in real time, with no need for painful, invasive repeat biopsies.
• Cost savings: QUS is far cheaper than repeated CT or MRI scans, making advanced survival prediction accessible to more patients.

Current Limitations and Future Research

While results are promising, the technology still has hurdles to clear before widespread adoption:

• Most validation studies have used small, single-institution datasets. Larger, more diverse multinational studies are needed to confirm accuracy across different patient populations.
• QUS imaging protocols vary between clinics, making it hard to standardize models. Industry-wide protocol standardization will be critical for broad use.

Future research will focus on combining QUS-ML predictions with genomic data and electronic health record information to deliver even more precise survival estimates.

Researchers are also working on embedding ML models directly into ultrasound machine software, so clinicians can get real-time survival risk predictions during patient scans.

Conclusion

Machine learning models trained on quantitative ultrasound imaging represent a major step forward in caring for locally advanced breast cancer patients receiving neoadjuvant chemotherapy.

By delivering accurate, non-invasive 3-year survival predictions, this technology empowers clinicians to make more informed treatment decisions and gives patients clearer expectations for their care journey.

As validation studies expand and tools become more accessible, ML-QUS could soon become a standard part of breast cancer care worldwide.