Draft Layout
Abstract:
In this body of work we apply convolutional neural networks to chromosome scans, for the purpose of detecting overlapping chromosomes in cytogenetic experiments. Our work is three fold, we introduce a technique to synthesize such data with labels. A new dataset with benchmarks, and lastly introduce a new state of the art model for such task. We use techniques that are heavily inspired by U-Net’s data augmentation to use the available annotated samples.
Outline:
- Introduction to the problem in cytogenetics
- Contributions:
- How to create the dataset.
- Benchmarks for the dataset with existing models, such as U-Net, DenseNet, FC-DenseNet (Hundred Layers of Tiramisu)
- Introducing the state of the art model for such task.
- Network Architecture
- Training Details
- Experiments
- Conclusion