Don't miss our weekly PhD newsletter | Sign up now Don't miss our weekly PhD newsletter | Sign up now

  Advancing Deep Learning (DL) Techniques for Medical Image Analysis


   Faculty of Engineering, Computing and the Environment

   Applications accepted all year round  Self-Funded PhD Students Only

About the Project

Introduction:

 

Lately, convolutional neural networks (CNNs) have demonstrated competitive performance in medical image analysis tasks, such as disease classification, tumor segmentation, and lesion detection [1]. CNNs have the ability of extraction of local features of the images. Lately, transformers have been employed for medical imaging (MI) analysis as a complete-stack clinical application. These include image reconstruction/synthesis, registration, segmentation, detection, and diagnosis. In comparison to the natural images, multi-modal medical images or multiplexed images have significant and explicit long-range dependencies that can enhance the performance of DL models. The rapid advancement in medical imaging strategies (Computed Tomography (CT), Single Photon Emission Computed Tomography (SPECT)), Positron Emission Tomography (PET), and Magnetic Resonance Imaging (MRI)) has drawn attention to fusing different modalities to aid experts' decision-making during the aided-diagnosis pipeline [2-3].

Need of the Study:

Current transformer-based architectures demand large-scale datasets to attain better performance. Nonetheless, MI datasets are moderately small making it challenging to use pure transformers to MI analysis.

Exploring preprocessing effect i.e., data augmentation and dimensional reductional effects on the performance.

Exploring property of the transformer effectiveness in extracting the relationship between sequences.

Exploration as an Open Topic:

A cascaded approach of CNN and multi-head self-attention (MHSA) approach for the classification of multi-modal medical images. The CNN will capture the local features and it will be fed to the MHSA for cross modality high level connections. The MHSA is a feature aggregation model and the local features from the CNN model are aggregated and process further for global feature extraction.


Computer Science (8) Medicine (26)

Register your interest for this project


Search Suggestions
Search suggestions

Based on your current searches we recommend the following search filters.