| KEYWORD |
Domain expert-inspired segmentation deep learning algorithms on medical images
keywords DEEP LEARNING, DEEP NEURAL NETWORKS, MACHINE LEARNING, MACHINE LEARNING, ARTIFICIAL NEURAL NETWORKS, MEDICAL IMAGING
Reference persons DANIELE APILETTI
Research Groups DAUIN - GR-04 - DATABASE AND DATA MINING GROUP - DBDM
Thesis type ANALISI DATI, ANALITICA E SPERIMENTALE, SVILUPPO SOFTWARE
Description Autosomal dominant polycystic kidney disease (ADPKD) is a monogenic, rare disease that affects about 1 in 1000 people worldwide and is characterized by the formation of multiple cysts that grow out of the renal tissues.
Image analysis of kidney tubule images is a promising tool for studying ADPKD and developing new therapeutic strategies. Nevertheless, this task is challenging due to several factors, such as the variability of cyst sizes, shapes, numbers, and distributions; the complexity of kidney anatomy and morphology; the low contrast and resolution of some imaging modalities; and the scarcity and cost of annotated data. Present-day methods rely primarily on semantic segmentation techniques, which aim to detect and segment each cyst in an image. However, these methods have some limitations and challenges, such as:
- They are standard implementations that do not incorporate domain knowledge into the segmentation process, such as prior information about cyst sizes, shapes, and locations.
- They lack interpretability and explainability of their segmentation results, such as providing confidence scores or rationale for their decisions.
The thesis aims to develop a novel framework for image segmentation of cysts on kidney tubules images that can incorporate domain-knowledge insights and mimic human annotator behavior to make more valuable predictions. This algorithm will then be evaluated on a real medical dataset confidentially shared by Istituto Mario Negri of Bergamo and compared with state-of-the-art solutions. Further developments will cover the generalizability of the developed approaches on related tasks, following the context of theory-guided data science.
Prerequisites: Strong background in computer science and good Python programming skills. Knowledge of machine learning, deep learning, and image analysis would also be beneficial. Additionally, experience with relevant tools and libraries for implementing machine learning algorithms (e.g., PyTorch) would be highly appreciated.
Deadline 14/06/2024
PROPONI LA TUA CANDIDATURA