Wykaz obszarów badawczych związanych z tagiem Deep-learning:
| # | Obszar badawczy | Dziedzina naukowa |
|---|---|---|
| 1 |
Self-supervised learning (SSL) methods have shown significant potential in improving the sample efficiency of deep learning and in providing a starting point for the learning of downstream tasks. Contrastive SSL methods have become a standard pre-training approach for a range of domains such as NLP and vision. Initial results suggest that non-contrastive SSL has been able to narrow the gap to the performance levels of contrastive methods, while obviating the need for explicit construction of negative samples. The aim of this research project is two-fold: 1) investigate the extent to which contrastive and non-contrastive methods can be used in novel SSL architectures, 2) examine whether the alignment of representations in SSL can be achieved by alternative methods, such as enforcing the ability to predict an input representation from the representations of similar inputs.
|
|
| 2 |
A prevalent theme present in the contemporary representation learning approaches is to pre-train large foundation models on huge datasets. Such approaches utilize static datasets constructed at a particular point in time, which contrasts with the constantly changing and expanding nature of data available on the internet. The proposed research will explore a new paradigm where the training dataset is constructed on the fly by querying the internet, enabling efficient adaptation of representation learning models to selected target tasks. The aims of this research project include 1) design methods to query relevant training data and use it to adapt the representation learning model in a continuous manner, 2) make progress towards building self-supervised methods that given a description of a task, autonomously formulate their learning curricula, query the internet for relevant training data, and use it to iteratively optimize the model.
|
|
| 3 |
Application of artificial intelligence methods in processing medical imaging data, analysis of histopathological data, and computer vision.
|
|
| 4 |
Application of artificial intelligence methods in processing medical imaging data, analysis of histopathological data, and computer vision.
|
|
| 5 |
Synthetic Data Generation for Machine Learning-Based Computer-Aided Diagnosis in Spinal Computed Tomography Machine learning-based computer-aided diagnosis in computed tomography (CT) is often limited by the scarcity of large, well-annotated clinical datasets. Acquiring representative CT examinations of vertebral fractures with expert annotations is particularly challenging because many clinically important fracture types occur infrequently, leading to severe class imbalance and reduced model generalization. This doctoral research aims to develop and evaluate methodologies for generating realistic synthetic three-dimensional CT data to support automated vertebral fracture classification. The work will focus on supervised deep learning models implementing the nine-class AO Spine fracture classification system. The research will investigate multiple approaches to synthetic volumetric CT generation, including generative artificial intelligence models and anatomically constrained non-rigid transformations derived from clinical CT examinations. A reference dataset of expert-annotated clinical CT scans will be used for model development, parameter optimization, and independent validation. A central objective is to quantify the effect of different synthetic data generation strategies on the performance, robustness, calibration, and generalization of deep learning models. Classifiers trained on real, synthetic, and hybrid datasets will be systematically compared using independent clinical CT examinations excluded from the training process. Particular attention will be given to improving the recognition of underrepresented fracture classes while maintaining high diagnostic performance across all AO Spine categories. The expected scientific contribution is the development of methodologies for generating anatomically realistic synthetic CT data and a comprehensive evaluation of their effectiveness in mitigating data scarcity in machine learning for computer-aided diagnosis.The research is expected to establish methodological principles for integrating synthetic data into medical imaging workflows and to improve the reliability and generalizability of automated vertebral fracture classification.
|
|
| 6 |
Development of Explainability Methods for Localizing Vertebral Fractures in Machine Learning-Based Computer-Aided Diagnosis Using Computed Tomography Machine learning-based computer-aided diagnosis in computed tomography (CT) has demonstrated promising performance in automated vertebral fracture classification. However, most supervised learning models are trained using only examination-level or vertebra-level labels and therefore provide little information about the image regions responsible for their predictions. This lack of interpretability limits clinical confidence and hinders the adoption of artificial intelligence in routine radiological practice. Pixel-level annotation of fracture regions could address this issue but is prohibitively time-consuming and expensive because it requires extensive expert involvement. This doctoral research aims to develop explainability methods capable of accurately localizing vertebral fractures using only classification labels during model training. The study will investigate whether explainability maps generated by deep learning classifiers can be transformed into reliable indicators of pathological regions without requiring manually annotated segmentation masks. The research will focus on automated classification of vertebral fractures according to the nine-class AO Spine classification system using three-dimensional CT data. Several deep learning architectures and explainability techniques will be investigated, including gradient-based attribution methods, activation map approaches, and perturbation-based explanations. Novel methodologies will be developed to improve the localization accuracy, anatomical consistency, and clinical interpretability of explainability maps. The proposed methods will be evaluated using expert-annotated clinical CT examinations, with particular emphasis on their ability to identify fracture regions while preserving high classification performance. The expected scientific contribution is the development of explainability methodologies that bridge the gap between image-level classification and lesion localization in medical imaging.
|