{"id":215,"date":"2026-01-13T10:53:24","date_gmt":"2026-01-13T09:53:24","guid":{"rendered":"https:\/\/vision.inf.uniri.hr\/?page_id=215"},"modified":"2026-01-20T15:09:28","modified_gmt":"2026-01-20T14:09:28","slug":"aicodd-project-info","status":"publish","type":"page","link":"https:\/\/vision.inf.uniri.hr\/?page_id=215","title":{"rendered":"AICODD – Project Info"},"content":{"rendered":"\n

Sponsor: <\/strong>Animal Eye Consultants of Iowa, Sinisa Grozdanic DVM, PhD, Dipl ACVO
Acronim: <\/strong>AICODD
Project Title: <\/strong>Image-based AI-assisted diagnostics of canine ocular disease – AICODD
Project Title (hr):
<\/strong>Dijagnostika o\u010dnih bolesti pasa uz pomo\u0107 umjetne inteligencije temeljena na slikama – AICODD
Project manager: <\/strong>prof. dr. sc. Marina Iva\u0161i\u0107 Kos prof. mat. i inf.
Project starts:<\/strong> 01.10.2022.
Duration: <\/strong>48 months<\/p>\n\n\n\n


<\/p>\n\n\n\n

Background:<\/strong>
Deep learning computer vision models have achieved excellent results in tasks such as object detection, image classification and segmentation, and is being applied in different domains to provide additional information and contribute to a better service.<\/p>\n\n\n\n

Work Plan:<\/strong>
Training of deep learning models in general and for specific tasks of eye detection, abnormality detection and disease classification requires a large dataset of images that are labeled by experts-veterinarians with appropriate annotations (i.e., specific diagnosis for disease classification, marking the affected region of the image for detection of abnormal growth in the eye for disease detection etc.).<\/p>\n\n\n\n

The labeled images will be collected continuously, however initial experiments with training deep learning models for eye detection and abnormal mass detection will begin as soon as an initial batch of labeled examples is gathered. For both tasks (eye detection, disease\/mass detection) first the collected labeled images will be prepared for use as a dataset for machine learning applications. This will include verification and cleanup of data, removing of unusable images from the set, normalization, image preprocessing and definition of training and testing subsets.<\/p>\n\n\n\n

After the initial datasets are prepared, preliminary tests will be conducted by training commonly used deep learning architectures for object detection on the datasets.<\/p>\n\n\n\n

Due to a large number of possible eye diseases and a great variety of possible appearances, the task of detecting the abnormal growths\/diseases will be defined according to the number and variety of available labeled images: if a smaller amount of data is available the task will be confined to the detection of presence\/no presence of disease\/growth, and if greater amount data is available, expanded to include classification of disease in different categories (granularity of which will also depend on available data).<\/p>\n\n\n\n

Based on the performance of the models trained in the preliminary tests on the test dataset, the best performing architectures for each task will be chosen as a foundation for further development of the models. The models based on the chosen architectures will subsequently be refined and optimized in several iterations to achieve the best detection score according to the standard metrics (e.g., mean average precision, mAP).<\/p>\n\n\n\n

Goals:<\/strong>
The aims of this proposal are:<\/p>\n\n\n\n