Subject

Prostate cancer is one of the most common cancer types among men worldwide. Magnetic Resonance Imaging (MRI) plays an important role in the detection of prostate cancer. Radiologists can identify lesions on the prostate MRI and assign a risk score to each lesion. However, the interpretation of the prostate MRI by radiologists is time-consuming and prone to inter-reader variability.
Deep learning models can be used to objectively predict prostate cancer from MRI scans. The goal of this thesis is to develop, train and evaluate such deep learning models.

Kind of work

Objectives
The aim of this thesis is to develop an autoencoder model that extracts features from prostate MRI scans. These features can be used for downstream tasks such as the prediction of prostate cancer.
The dataset of the PI-CAI (Prostate Imaging: Cancer AI) challenge will be used to train and test the deep learning models.

The objectives of the master thesis are:
1. Reviewing state-of-the-art autoencoder architectures in medical imaging.
2. Designing and implementing autoencoder models for feature extraction.
3. Predicting clinically significant prostate cancer with the extracted features.
4. Comparing the performance of different model architectures and configurations.

Description of Work
The project consists of the following tasks:
- Literature study
- Downloading and processing of the PI-CAI dataset
- Implementation of unsupervised autoencoder models for feature extraction
- Classification of features extracted in the previous task
- Evaluation and comparison of the performance of the autoencoder and classification models