Side Project: Segment Overlapping Chromosomes

Overview

The goal of this peoject is to identify and segment the overlapping chromosomes.

The model used in this task is the autoencoder with convolutional neural network structure.

The related information regarding this dataset can be accessed below:

Result

The predicted outcome on one of the example from test dataset:

The result looks not bad, it achieves:

  • Accuracy: 0.996
  • Precision: 0.75
  • Recall: 0.84
  • F1 score: 0.78

Further

  • Use some effective preprocessing techniques
  • Tune the structure to improve the performance on recall