Synthetic Data Augmentation for Plasmocyte Detection in Bone Marrow Images

Nome do aluno JORGE LUIS DA SILVA BATISTA FILHO
Título do trabalho Synthetic Data Augmentation for Plasmocyte Detection in Bone Marrow Images
Resumo do trabalho Accurate detection of plasmocytes in bone marrow microscopy images is a key step in the diagnosis of multiple myeloma, which relies on the plasmocyte-to-non-plasmocyte cell ratio. However, the limited availability of annotated medical data poses a challenge for training robust object detection models. In this work, we investigate the use of synthetic data augmentation to improve plasmocyte detection performance. We generate images of individual plasmocytes using multiple generative models, including diffusion models (U-Net-based), WGAN-GP, and Diffusion Transformers (DiT). These synthetic cells are then composited into real bone marrow microscopy images, with automatic generation of corresponding annotations. The resulting dataset is used to train a YOLO-based object detector, which is evaluated for plasmocyte detection. The research proposal focuses on designing a framework capable of creating synthetic cells to achieve two main purposes: augmenting data to improve performance in a clinically relevant diagnostic scenario, and supporting the training of specialists to identify plasma cells and reduce diagnostic errors.
Orientador Ricardo Araújo Rios
Membro Titular 1 (com afiliação) Danilo Barbosa Coimbra (UFBA)
Link para o curriculum lattes http://lattes.cnpq.br/9590398895954821
Membro Titular 2 (com afiliação) Rubisley de Paula Lemes (FBA)
Link para o curriculum lattes http://lattes.cnpq.br/4230260717556147
Suplente 1 (com afiliação) Ricardo Bastos Cavalcante Prudêncio (Universidade Federal de Pernambuco)
Link para o curriculum lattes http://lattes.cnpq.br/2984888073123287
Suplente 2 (com afiliação) Elaine Ribeiro de Faria (Universidade Federal de Uberlândia)
Link para o curriculum lattes http://buscatextual.cnpq.br/buscatextual/visualizacv.jsp?id=K4127008H7
Data do exame 10 Jun, 2026
Horário do exame 2:00 PM

 

Data da Defesa: 
10/06/2026 - 14:00
Tipo de Defesa: 
Qualificação de Mestrado