Are Your “High‑Accuracy” Models Built on Invisible Cracks?
Machine‑learning models that predict survival curves, tumor recurrence, or treatment response can look spectacular in a Jupyter notebook—right up until they meet real clinical data. The silent culprit? Information leakage: hidden shortcuts that slip illicit clues from the future or from duplicate patient records into the training set.
Dr Julia Sidorova (Research Scientist, Bioinformatics Platform, CIBER‑EHD, Spain) has made it her mission to reveal—and repair—those cracks. At Cancer 2025 she’ll dissect how leakage sneaks into pipelines, why it cripples external validation, and which safeguards every oncology data team should adopt.
What She’ll Cover on Stage
| Leakage Type | Real‑World Example | Fix You’ll Learn |
|---|---|---|
| Temporal leakage | Future lab results “accidentally” present during training | Time‑aware splits, forward‑chaining validation |
| Patient‑overlap leakage | Multiple CT slices from the same patient split across folds | Grouped k‑fold, patient‑level indexing |
| Pretraining leakage | Public backbone pre‑exposed to pathology labels | Audit of pretraining datasets & feature ablation |
| Site/batch leakage | Model learns the scanner brand, not the tumor | ComBat batch correction, domain adversarial nets |
Key takeaway: When Dr Sidorova removed leakage, deep nets in survival analysis “often under‑performed vanilla linear regression.” Let that sink in.
Meet Dr Julia Sidorova
-
PhD, Universidad Pompeu Fabra (Barcelona)
-
Research Scientist, Instituto Carlos III de Salud (Madrid)
-
Editorial Board, Frontiers in Neurology (Biomarkers) & IJMS (SI: AI in Molecular Cancer)
-
２０＋ peer‑reviewed papers merging classical stats with deep learning to test what really works
Event Snapshot
| Conference | International Conference on Cancer Science & Research (Cancer 2025) |
| Dates | 17 – 19 November 2025 |
| Format | Hybrid—Singapore onsite + fully interactive virtual platform |
Why Attend?
-
Bulletproof your ML pipelines before they hit the clinic
-
Collect CPD credits and SCOPUS‑indexed proceedings
-
Network with oncologists, bioinformaticians, and AI regulators in one forum
Register Now
Secure your seat or virtual pass → https://cancer.miconferences.com/register
Questions?
📧 cancer@mathewsconference.com | ☎ +1 312 462 4448 | 💬 WhatsApp https://wa.me/14243770967
Let’s build AI models that earn clinicians’ trust—starting with a leak‑tight foundation.
#InformationLeakage #ClinicalAI #OncologyData #SurvivalAnalysis #Bioinformatics #MachineLearning #Cancer2025 #SingaporeEvents #CancerResearch #ModelValidation #DeepLearning #HealthcareAI #HybridConference #SingaporeEvents #OncologyConferences #CancerEvents
