Subject

The last decades, many computer aided detection and diagnosis (CAD) systems have been proposed to diagnose breast cancer based only on the properties of microcalcifications (MCs) - the main indicators of an early breast cancer. Recently, promising results have been achieved by using handcrafted and/or deep learning features extracted from high-resolution 3D MCs images. In this thesis (continuation of previous work already performed), The focus is: (a) to evaluate (and quantify) if using transfer learning (TL) can help to get a better performance (b) quantify and analyze the computation efficiency when using and not TL.

Kind of work

- Literature review about: (a) the medical problem the student is trying to solve, (b) deep convolutional neural networks, (c) transfer learning.
- Get familiar with the dataset to be used (high resolution 3D micro-CT images).
- Evaluate the deep learning model performances when using and not TL, with/without offline data augmentations.
- Evaluate and analyse in depth the influence of using transfer learning to classify breast microcalcification.
- Writing and presentation.

Framework of the Thesis

- https://arxiv.org/abs/2004.07882

Number of Students

1