Ashwin Raju

Machine Learning Enthusiast

image

I am a Machine learning enthusiast who has 7 years of experience in applying computer vision, machine learning alogrithms to solve problems that directly impact people. sI have developed and patented algorithms which solves real world problems in medical field. I have expertise in applying machine learning models to diverse areas such as laparoscopic surgery, radiology, histology.


Work Experience

Research Intern

PAII Inc | Sept 2020 - Dec 2020

Designed a statistical shape model that can be incorporated with the deep learning framework to estimate rigid and non-rigid pose estimations. Our designed statistical shape model learns the prior anatomical shape and significantly reduces the variance of multi-organs. Tested on Liver and Larynx organs.

  • Published at AAAI'22 with Oral acceptance.

Research Intern

PAII Inc | Sept 2019 - Dec 2019

Designed a framework to integrate co-training, hetero-modality and adversarial learning to segment multi-organs in a semi-supervised approach. Improved the pathological liver mask dice coefficients by ranges of 4.2% to 9.4%.

  • Published at MICCAI'20.

Research Intern

PAII Inc | May 2019 - Sept 2019

Designed a framework to incorporate minimal-labor user interactions to segment multi-organs. Improved the pathological liver mask dice coefficients by 3% when compared to the state of the art models.

  • Published at ECCV'20.

Graduate Research Assistant

University of Texas, Arlington | 2017 - 2022

  • Designed a framework to classify Cancer stages from histology images by incorporating Attention multi-instance learning in a form of a graph.
  • Implemented several Deep learning based algorithms using tensorflow as a proof of concept.
  • Designed a framework to classify Cancer stages from histology images by incorporating Attention multi-instance learning in a form of a graph.

Projects

Cancer cell localization

Grant Project

Worked on a project which involves localization of 51 different cancer cells present in a gold standard dataset. Working on building a novel Semi-Supervised model keeping Feature Pyramid Network, Attention Model as baseline.

Nuclei cell segmentation

Open Source

Implemented several state-of-the-art models such as Mask-RCNN, UNet, SegNet, DeepLab v2 to understand the architecture and evaluate the performance of these models for Nuclei Cell Segmentation. Experimented the state-of-the-art techniques with several public datasets.

Multi-class artefact detection in video endoscopy

Open Source

Classification of 7 classes in Endoscopic videos. The model is built on ensembling VGGNet and GoogLeNet to produce high performance results. The network achieves a Mean Average Precision of 63.8% in test dataset and won first place in M2CAI challenge.

Publications

Interests

Travelling, Binge watching