Deep Learning for Ovarian Cancer Survival Prediction
Author Information
Author(s): Spirina Menand, Elena Conceptualization, Manon De Vries-Brilland, Leslie Tessier, Jonathan Dauvé, Mario Campone, Verrièle Véronique, Nisrine Jrad, Jean-Marie Marion, Pierre Chauvet, Christophe Passot, Alain Morel, Roberto Piergentili
Primary Institution: Laboratoire Angevin de Recherche en Ingénierie des Systèmes, Université d’Angers, France
Hypothesis
Can deep learning models effectively predict survival outcomes in ovarian cancer patients using transcriptomic data?
Conclusion
Deep learning models can identify molecular pathways associated with survival in ovarian cancer patients, but their direct application in clinical practice remains challenging.
Supporting Evidence
- The model stratified TCGA patients into high-risk and low-risk groups based on identified molecular pathways.
- Validation on an in-house dataset showed a p-value of 0.229, indicating no significant difference in survival.
- External validation on the JGOG-OV dataset showed a significant survival difference with a p-value of 0.006.
Takeaway
Researchers used computer models to look at gene data from ovarian cancer patients to help predict who might live longer. They found some important gene patterns that could help doctors make better treatment choices.
Methodology
The study used TCGA ovarian cancer transcriptomic data from 372 patients and an in-house validation dataset of 12 patients to train and evaluate deep learning survival models.
Potential Biases
Potential overfitting due to high-dimensional data and small validation sample size.
Limitations
The model did not generalize well to new unseen data, particularly in the in-house validation dataset.
Participant Demographics
The study included 372 ovarian cancer patients from TCGA and 12 patients from an in-house dataset, primarily diagnosed with high-grade serous ovarian carcinoma.
Statistical Information
P-Value
0.025
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website