We are pleased to announce our network study Deep Learning Comparison.
Study leads: Henrik John (@lhjohn), Chungsoo Kim (@Chungsoo_Kim), Jenna Reps (@jennareps), and Egill Fridgeirsson (@egillax)
GitHub: Deep Learning Comparison - GitHub Repository
Protocol: Deep Learning Comparison - Protocol
Infrastructure: To execute the analysis an Nvidia GPU with CUDA support is required. We recommend a minimum of 12 GB video memory; more is preferred to speed up analysis.
Participant deadline: Please let us know before 1 March, if you are interested in joining the study.
Aim: Assess the value of deep learning methods over conventional methods for the development of clinical prediction models. The specific diseases under consideration are dementia in individuals over 55, lung cancer in those over 45, and bipolar disorder in patients misdiagnosed with major depressive disorder.
Rationale: Deep learning techniques have proven to be highly effective for prediction on unstructured data, such as image and text. However, when applied to structured, sparse, and high-dimensional healthcare data deep learning often yields results comparable to those of simpler, conventional prediction methods. In this study we develop and validate clinical prediction models using deep learning and conventional approaches to compare their discriminatory power and calibration on OMOP CDM data.