Impact of Data Quality and Quantity on its Effectiveness on Multi-Stage Transfer Learning Using MRI Medical Images

Presentation Type

Poster

Student

Yes

Track

Health Care Application

Abstract

Multi-Stage Transfer Learning (MSTL) has been becoming a very promising area of research in the field of medical imaging. Model architecture based on Multi-Stage TL provided promising results surpassing the previous standards. In our study, we provide an overview of Multi-Stage TL and its implementation in medical imaging followed by reviewing the research work in the field of transfer learning in medical imaging. Our objective is to investigate and understand the different effects of data quality and quantity on Multi-Stage Transfer learning using the MRI images. We propose an MSTL model comprises of 4 different stages, in the first stage the model adapts the features and weights from a pre-trained network, second stage will include the domain adaptation having similar domain data with previous weights being fine-tuned, third stage is split into 3 separate layers each investigating the impact of Data Quality, Quantity and image features. In the final stage, we will apply the weights learned from the previous stages into the completely new dataset (Target/Problem area) and analyze its effects. Our study discusses the utilization of Multi-Stage transfer in medical imaging using the CNN architectures such as Inception V-3, AlexNet, and ResNet and investigate the current challenges in medical imaging domain such as computational complexity, domain adaptation and effectiveness of data quality and quantity using TL and Multi-Stage TL and proposed the future research areas.

Start Date

2-11-2020 1:00 PM

This document is currently not available here.

Share

COinS
 
Feb 11th, 1:00 PM

Impact of Data Quality and Quantity on its Effectiveness on Multi-Stage Transfer Learning Using MRI Medical Images

Volstorff A

Multi-Stage Transfer Learning (MSTL) has been becoming a very promising area of research in the field of medical imaging. Model architecture based on Multi-Stage TL provided promising results surpassing the previous standards. In our study, we provide an overview of Multi-Stage TL and its implementation in medical imaging followed by reviewing the research work in the field of transfer learning in medical imaging. Our objective is to investigate and understand the different effects of data quality and quantity on Multi-Stage Transfer learning using the MRI images. We propose an MSTL model comprises of 4 different stages, in the first stage the model adapts the features and weights from a pre-trained network, second stage will include the domain adaptation having similar domain data with previous weights being fine-tuned, third stage is split into 3 separate layers each investigating the impact of Data Quality, Quantity and image features. In the final stage, we will apply the weights learned from the previous stages into the completely new dataset (Target/Problem area) and analyze its effects. Our study discusses the utilization of Multi-Stage transfer in medical imaging using the CNN architectures such as Inception V-3, AlexNet, and ResNet and investigate the current challenges in medical imaging domain such as computational complexity, domain adaptation and effectiveness of data quality and quantity using TL and Multi-Stage TL and proposed the future research areas.