Projects

Quality control of brain tumor segmentations

Deep learning models have been extensively developed for brain tumor segmentations from MRI images, mostly in a supervised way where ground truth labels are available. However, this condition is not warranted in datasets that do not have ground truth annotations. In this work, we developed a novel method that predicted the Dice score of the tumor segmentations without the ground truth, while also generating the error map that highlighted the areas where the predicted segmentation and the ground truth disagreed.

Google Summer of Code

A weekly notebook of my coding journey with GSoC 2023

Deep learning for transcription factors binding sites predictions

In this work I am investigating novel multi-modality deep learning models for more accurate transcription factors binding sites predictions

Testing and maintenance of Nextflow pipeline for identifying genetic risk loci

I was responsible for testing a Nextflow pipeline for identifying genetic risk loci enriched at open chromatin regions. The tool included the ENCODE pipeline for ATAC-seq and ChIP-seq, HOMER for motif enrichment analysis, and RELI for genetic risk loci identification.

Realistic generation of synthetic pediatric liver ultrasound images for liver disease classification

Large-scale labeled biomedical datasets are difficult to access, and these datasets remain small in the medical imaging domain compared to other common computer visiom datasets. Therefore, we presented a method of using different models of generative adversarial networks to generate synthetic liver ultrasound images. Our results showed that the synthetic images had good quality compared with the real images, and this synthetic dataset would be promising as a data augmentation method for liver disease classification task.