Can a Simple Photo Help Predict Cancer Survival Rates? FaceAge AI
In this article
Facial Age: A Deep Learning System in Cancer Diagnosis
On May 8, 2025, The Lancet Digital Healthpublished a study showing that FaceAge , a deep learning system, estimates biological age from facial photographs and, in cancer patients, improves survival prediction .
Although “first impressions” play a role in physicians’ clinical decisions, they remain subjective and provide only a rough idea of biological age . A qualitative and standardized measure was needed. FaceAge aims to fill this gap with a quantitative “facial age” derived from a single facial photograph .
Summary: In facial photographs, cancer patients appear approximately 5 years older on average than people without cancer. Facial age is independently associated with survival across multiple cancer groups . In palliative care, using AI improves the accuracy of physicians’ predictions about a patient’s course by approximately 6%. In other words, AI support increases doctors’ ability to predict correctly from 74% to 80%.
📷 Clinical Decision-Making With Photographs and Facial Age
FaceAgeuses a facial photograph taken with a standard webcam/phone to estimate age, adding an objective layer to the physician’s subjective performance assessment. Patients whose facial age is younger than their chronological age tend to have better outcomes after treatment.
🧠 How Was the Model Trained?
- Training: IMDb–Wiki database provided 56,304 facial images.
- Technical validation: UTKFace dataset provided 2,547 images.
🏥 Clinical Testing (US & Netherlands)
- Total: 6,196 cancer patients (facial photographs taken at the start of radiotherapy).
- Reference: A cohort of 535 people without cancer.
- Analysis: Kaplan–Meier, Cox regression; multivariable adjustments.
Image description: The FaceAge algorithm uses deep learning to estimate biological age from a facial photograph and investigate survival prediction in cancer patients. The study compared discovery data from 58,851 healthy individuals with clinical data from 6,196 cancer patients. The findings suggest that looking older than one’s actual age may be associated with poorer survival.
📊 Study Results (Key Numbers)
Image Description: The AI-powered FaceAge algorithmwas evaluated as a powerful tool for predicting survival by estimating biological age from cancer patients’ facial photographs. Analyses identified three main groups:
- Green Group: Patients whose faces looked younger than their chronological age – median survival 40 months.
- Blue Group: Patients whose facial age equaled their chronological age (reference group) – median survival 28 months.
- Red Group: Patients whose faces looked older than their actual age – median survival of only 18 months.
Hazard ratio (HR) analyses also supported this difference. Younger-looking patients showed lower risk, while older-looking patients showed higher risk . The black line represents these HR values.
🔹 Age difference: On facial analysis, cancer patients appeared an average of 4.7 years older than healthy people of the same age (P<0.0001).
🔹 Survival association (FaceAge ↑ → risk ↑):
- All cancer types combined: Risk of death: 15% increase (HR 1.151; P=0.013)
- Thoracic (lung/chest) cancers: Risk of death: 11% increase (HR 1.117; P=0.021)
- Palliative care patients: Risk of death: 11% increase (HR 1.117; P=0.021)
🔹 Clinical benefit: In palliative care, physicians’ accuracy in predicting 6-month survival increased from 74% to 80% . This represents a 6% absolute improvement (P<0.0001).
👉 In summary: Looking older than one’s actual age is associated with a poorer prognosis in cancer. The FaceAge algorithm captures this signal of biological aging, substantially improving physicians’ survival predictions.
🧬 Molecular Analysis
CDK6 (Cyclin-Dependent Kinase 6)is an enzyme that regulates the cell cycle and plays a critical role in cell proliferation and aging. The study found an inverse relationship between FaceAge and CDK6 expression :
- In patients whose faces looked older than their actual age, CDK6 levels were low .
- This finding suggests accelerated cellular aging.
What does this mean?
- Facial age may be not only an aesthetic measure, but also a molecular biomarker of aging .
- Chronological age → indicates the number of calendar years.
- Molecular age → reflects how much cells, genes and proteins have “aged.”
- FaceAge → may capture the outward manifestation of this molecular aging (facial cues).
Therefore, the facial age estimated by FaceAge and its negative correlation with CDK6 levelsshow that AI-based facial analysis is meaningful not only clinically, but also at the biological and molecular levels.
📌 At a Glance: Methodology and Impact
| Component | Summary |
|---|---|
| Training Data | IMDb–Wiki (56,304 images) → UTKFace (2,547) technical validation |
| Clinical Cohort | A total of 6,196 cancer patients; reference group of 535 cancer-free individuals |
| Main Result | Facial age ↑ → poorer survival; in palliative care, AUC 0.74 → 0.80 |
| Molecular Findings | Inverse relationship with CDK6; no significant relationship with chronological age (after multiple adjustments) |
🔍 Clinical Significance
- Makes assessment objective: Adds to the physician’s subjective first impression a standard metric .
- Improves stratification: Provides independent prognostic information across multiple cancer groups.
- Provides decision support: Particularly in palliative treatment planning, it improves predictive performance.
⚠️ Limitations & Ethical Concerns
- Training biases: Overrepresentation of celebrities in IMDb–Wiki and possible cosmetic/digital alterations.
- Generalizability: Need for validation in larger and more diverse clinical cohorts.
- Risk of misuse: Inappropriate use of facial age in insurance and similar fields requires a regulatory framework .
📝 Conclusion
FaceAge has the potential to use an easily accessible facial photograph to produce a clinically meaningful biomarker. With larger multicenter validation, it could be safely integrated into clinical decision support workflows.
Sources
Bontempi, Dennis et al.
The Lancet Digital Health, Volume 7, Issue 6, 100870
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.