Název:	The Disease of the Canine Eye - From Image to Diagnosis Using AI
Autoři:	Buric, Matija Ivasic-Kos, Marina Martincic-Ipsic, Sanda
Citace zdrojového dokumentu:	WSCG 2024: full papers proceedings: 32. International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, p. 377-380.
Datum vydání:	2024
Nakladatel:	Václav Skala - UNION Agency
Typ dokumentu:	konferenční příspěvek conferenceObject
URI:	http://hdl.handle.net/11025/57412
ISSN:	2464–4625 (online) 2464–4617 (print)
Klíčová slova:	počítačové vidění;velké jazykové modely;segmentace obrazu;architektura U-Net;veterinární oftalmologie;lokalizace onemocnění;diagnostické nástroje;veterinární diagnostické zobrazování;automatizovaná diagnóza
Klíčová slova v dalším jazyce:	computer vision;large language models;image segmentation;U-Net architecture;veterinary ophthalmology;disease localization;diagnostic tools;veterinary diagnostic imaging;automated medical diagnosis
Abstrakt v dalším jazyce:	This research examines the application of computer vision (CV) and large language models (LLM) in diagnosing eye diseases in dogs. The study utilizes a U-Net framework, incorporating convolutional neural networks (CNNs) such as ResNet, Inception, VGG, and EfficientNet, to enhance the segmentation of eye disease areas. Along the base U-Net model, four U-Net-based models were developed and evaluated on a dataset specifically generated for this purpose, classifying eye diseases into four categories. The performance of the enhanced U-Net architectures was found to be superior to that of the standard U-Net, with the U-Net modified with ResNet34 achieving the best segmentation accuracy, as measured by a Jaccard index of 66.6% on a custom test set. The segmented images were then diagnosed using various LLMs, including ChatGPT, Mistral, Gemini (Bard), Claude, and Llama-2, which were assessed using 15 different symptom sets. The study demonstrates that combining advanced image segmentation techniques with LLMs can improve diagnostic accuracy in veterinary medicine. The approach leverages the segmentation capabilities of U-Net for precise localization and the diagnostic ability of LLMs to interpret symptoms, facilitating enhanced diagnostic tools. This method could be applicable to other medical diagnostic areas requiring similar dual capabilities.
Práva:	© Václav Skala - UNION Agency
Vyskytuje se v kolekcích:	WSCG 2024: Full Papers Proceedings

Všechny záznamy v DSpace jsou chráněny autorskými právy, všechna práva vyhrazena.