Detection, segmentation, and classification of pathologies across colonoscopy, capsule endoscopy, chest radiography, MRI, CT, ophthalmology, histopathology, and PET scans. Research covers strabismus, glaucoma, AMD, colon and penile cancer, kidney tumors, cardiac structures, endometriosis, and seismic gas segmentation.
The Computer Vision & Medical Imaging research line is the largest and most internationally recognized at NCA-UFMA. Two researchers from this group rank among the top 5 scientists globally in AI for Strabismus publications.
Ophthalmology: Strabismus detection via the Bruckner test, glaucoma diagnosis from fundus photographs using NEAT-based genetic architecture search (ALAN system), age-related macular degeneration (AMD) layer segmentation in OCT images, dry eye syndrome detection, oculoplastic pathology classification (ptosis, dermatochalasis, entropion, ectropion, pterygium).
Oncology: Colon cancer and penile cancer histopathological classification via transfer learning and optimization; melanoma segmentation from dermoscopy; whole-body PET scan cancer detection; breast cancer classification from ultrasound; multi-label chest X-ray diagnosis (14 labels, CheXpert dataset, 224,316 exams).
Gastrointestinal: Polyp detection in colonoscopy using RetinaNet with Feature Pyramid Network; pathology detection in capsule endoscopy.
Pulmonary: Pneumonia and COVID-19 classification from CT using Vision Transformer; tuberculosis detection from chest X-ray.
CT-based automatic segmentation of liver, kidneys, pancreas, heart, and spinal cord; kidney tumor segmentation from the KiTS19 dataset; short-axis cine-MRI cardiac structure segmentation; endometriosis segmentation from MRI; coronary calcification quantification from CT; synthetic contrast-enhanced CT generation from non-contrast CT using GANs.
Application of natural language modeling to seismic trace tokenization (Tracepiece) for natural gas segmentation in seismic reflection images, enabling automated reservoir characterization.
CEMAR-CELPA / ANEEL P&D
Mobile image acquisition and computational intelligence pipeline for reading and validating residential electricity meter digits. Uses FAST/GFTT feature detection for meter identification, CNN digit segmentation, and an automated audit trail.
Equatorial Energia / ANEEL P&D
Self-reading system enabling residential consumers to submit meter readings via WhatsApp chatbot (AutoClara virtual assistant). Incorporates periodic self-audit with image acquisition to reduce fraud, decrease operational costs, and improve customer satisfaction.