Titulo Estágio
Explainable AI-Driven Diagnosis of Fungal Infections using a Biological Image Dataset
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
CISUC
Enquadramento
Invasive fungal infections represent a critical global health challenge with 1.5 million lives lost annually. Fungal infections are among the most common hospital-acquired infections with high mortality rates and difficult diagnostic procedures. The traditional diagnostic techniques, such as blood culturing suffer from long processing times, low sensitivity, and the need for specialized equipment. To address these challenges, this project proposes the creation of a new dataset based on fungal images to develop AI-based diagnostic tools for effective identification of fungal species using microscopy images. The project will explore Vision Transformers (ViTs) and convolutional neural networks (CNNs), powerful deep learning models designed for image recognition, trained on the dataset to be created. The goal is to create an accessible dataset and explainable AI-driven solution for fungal prediction applicable in scenarios where traditional methods are unavailable.
Objetivo
The main objective of this project is to organize and pre-process fungal image data that has already been acquired and annotated to create a new dataset to train and test artificial intelligence methods to perform explainable fungal predictions. Hence, the project aims to develop and optimize deep learning (DL) models, specifically Vision Transformers (ViT) and Convolutional Neural Networks (CNN) to achieve accurate and robust predictions of fungal infections. A key goal is to ensure the predictions are explainable and show their potential for real-time diagnostics in clinical environments. Finally, the project seeks to make a scientific contribution by publishing a research paper and the processed dataset for AI-driven fungal predictions.
Plano de Trabalhos - Semestre 1
1. Literature review on computational and AI-based prediction methods using biological image data.
2. Organize and begin the processing of annotated fungal images already acquired to build a fungal image dataset.
3. Write an intermediate report summarizing the progress, challenges, and initial findings.
Plano de Trabalhos - Semestre 2
1. Apply data preprocessing techniques to improve the dataset for better AI and DL model generalization.
2. Train Vision Transformer (ViT) models (using pre-trained networks when applicable) and compare their performance with baseline CNN models for fungal classification and prediction.
3. Evaluate model performance using key metrics such as accuracy, precision, recall, and F1-score.
4. Integrate explainability techniques to interpret and visualize predictions and provide more transparent/ interpretable results.
5. Prepare and submit a research paper to a scientific journal or conference.
Condições
The candidate must be strongly motivated and demonstrate solid research skills, particularly in scientific literature review, experimental design, data collection, data curation. Familiarity with Python, artificial intelligence and machine learning libraries (e.g., PyTorch, TensorFlow) is beneficial, however, the focus will be on building a high-quality dataset, develop and evaluate deep learning models, and scientific reporting.
The models developed and fungal image dataset created may be published as open-access research.
Orientador
Bernardete Ribeiro
bribeiro@dei.uc.pt 📩