Luísa Vieira
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Luísa Vieira has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- Bernardo Gonçalves
Physics Department, NOVA School of Science and Technology, 2829-516 Caparica, Portugal
2 shared publications - Pedro Vieira
Department of Physics, Faculty of Science and Technology, NOVA University of Lisbon, Caparica Campus, 2829-516 Caparica, Portugal
2 shared publications - Gonçalo Saldanha
Hospital Garcia de Orta, 2805-267 Almada, Portugal
1 shared publication - Mariana Silva
Physics Department, NOVA School of Science and Technology, 2829-516 Caparica, Portugal
1 shared publication - Miguel Ramalho
Department of Radiology, Hospital da Luz, 1500-650 Lisbon, Portugal
1 shared publication
Latest Publications
- Open Access
- 8 April 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (8) , 4100
A Machine Learning Pipeline for Adenoma Detection in MRI: Integrating Deep Learning and Ensemble Classification
Adrenal lesions are common findings in abdominal imaging, with adrenal adenomas being the most frequent type. Accurate detection of adrenal adenomas is essential to avoid unnecessary diagnostic procedures and treatments. However, conventional imaging-based evaluation relies heavily on the expertise of radiologists and can be complicated by pseudo-lesions, overlapping imaging features, and suboptimal imaging techniques. To address these challenges, we propose an end-to-end machine learning pipeline that integrates deep learning-based lesion detection (FCOS) with an ensemble classifier for adrenal lesion classification in MRI. Our pipeline operates directly on broader regions of interest, eliminating the need for manual lesion segmentation. Our method was evaluated on a multi-sequence MRI dataset comprising 206 adenomas and 45 non-adenomas. The pipeline achieved 87.45% accuracy, 87.33% specificity, and 87.63% recall for adenoma classification, demonstrating competitive performance compared to prior studies. The results highlight strong non-adenoma identification while maintaining robust adenoma detection. Future research should focus on dataset expansion, external validation, and comparison with radiologist performance to further validate clinical applicability.
- 8 April 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (8) , 4100
Open AccessA Machine Learning Pipeline for Adenoma Detection in MRI: Integrating Deep Learning and Ensemble Classification
Adrenal lesions are common findings in abdominal imaging, with adrenal adenomas being the most frequent type. Accurate detection of adrenal adenomas is essential to avoid unnecessary diagnostic procedures and treatments. However, conventional imaging-based evaluation relies heavily on the expertise of radiologists and can be complicated by pseudo-lesions, overlapping imaging features, and suboptimal imaging techniques. To address these challenges, we propose an end-to-end machine learning pipeline that integrates deep learning-based lesion detection (FCOS) with an ensemble classifier for adrenal lesion classification in MRI. Our pipeline operates directly on broader regions of interest, eliminating the need for manual lesion segmentation. Our method was evaluated on a multi-sequence MRI dataset comprising 206 adenomas and 45 non-adenomas. The pipeline achieved 87.45% accuracy, 87.33% specificity, and 87.63% recall for adenoma classification, demonstrating competitive performance compared to prior studies. The results highlight strong non-adenoma identification while maintaining robust adenoma detection. Future research should focus on dataset expansion, external validation, and comparison with radiologist performance to further validate clinical applicability.
- Open Access
- 7 June 2024
- Published by MDPI AG in BioMedInformatics
- Vol. 4 (2) , 1506-1518
Abdominal MRI Unconditional Synthesis with Medical Assessment
Current computer vision models require a significant amount of annotated data to improve their performance in a particular task. However, obtaining the required annotated data is challenging, especially in medicine. Hence, data augmentation techniques play a crucial role. In recent years, generative models have been used to create artificial medical images, which have shown promising results. This study aimed to use a state-of-the-art generative model, StyleGAN3, to generate realistic synthetic abdominal magnetic resonance images. These images will be evaluated using quantitative metrics and qualitative assessments by medical professionals. For this purpose, an abdominal MRI dataset acquired at Garcia da Horta Hospital in Almada, Portugal, was used. A subset containing only axial gadolinium-enhanced slices was used to train the model. The obtained Fréchet inception distance value (12.89) aligned with the state of the art, and a medical expert confirmed the significant realism and quality of the images. However, specific issues were identified in the generated images, such as texture variations, visual artefacts and anatomical inconsistencies. Despite these, this work demonstrated that StyleGAN3 is a viable solution to synthesise realistic medical imaging data, particularly in abdominal imaging.
- 7 June 2024
- Published by MDPI AG in BioMedInformatics
- Vol. 4 (2) , 1506-1518
Open AccessAbdominal MRI Unconditional Synthesis with Medical Assessment
Current computer vision models require a significant amount of annotated data to improve their performance in a particular task. However, obtaining the required annotated data is challenging, especially in medicine. Hence, data augmentation techniques play a crucial role. In recent years, generative models have been used to create artificial medical images, which have shown promising results. This study aimed to use a state-of-the-art generative model, StyleGAN3, to generate realistic synthetic abdominal magnetic resonance images. These images will be evaluated using quantitative metrics and qualitative assessments by medical professionals. For this purpose, an abdominal MRI dataset acquired at Garcia da Horta Hospital in Almada, Portugal, was used. A subset containing only axial gadolinium-enhanced slices was used to train the model. The obtained Fréchet inception distance value (12.89) aligned with the state of the art, and a medical expert confirmed the significant realism and quality of the images. However, specific issues were identified in the generated images, such as texture variations, visual artefacts and anatomical inconsistencies. Despite these, this work demonstrated that StyleGAN3 is a viable solution to synthesise realistic medical imaging data, particularly in abdominal imaging.