VN October 2025

Vetnuus | October 2025 23 4.2 Metrics The evaluation of the experiment is centered around two primary aspects: the segmentation of symptoms and the interpretation of input images. The results from both stages of the experiment will be quantitatively assessed using standard metrics for segmentation accuracy and text similarity, comparing the model outputs to ground truth data. The following subsections provide detailed explanations of the metrics and procedures employed for evaluating model performance. 4.2.1 Segmentation metrics To assess the effectiveness of the segmentation models, several well-established metrics are applied. These metrics offer a robust evaluation of how accurately the models segment the key symptoms in canine ocular images. • Jaccard Index (Intersection-over-Union, IoU): The Jaccard Index measures the overlap between the predicted segmentation and the ground truth, reflecting how well the model can identify the same regions as the ground truth segmentation. It is calculated as (11): Where TP represents true positives (correctly identified regions), FP represents false positives (incorrectly identified regions), and FN represents false negatives (missed regions) as described in Figure 21. IoU provides a direct measure of how much of the predicted segmentation overlaps with the actual, intended segmentation. • Dice Index: A measure similar to the Jaccard Index but with greater sensitivity to overlap, the Dice Index evaluates the degree of similarity between the predicted and actual segmentations. It is calculated using the following formula (12): Leading Article Figure 20. The test dataset contains 25 samples, evenly distributed across five different diseases. Symptom 1 represents a colored mass protruding in the corner of the eye, Symptom 2 refers to cloudiness or haziness in the cornea, Symptom 3 signifies redness of the sclera, and Symptom 4 denotes excessive tearing 11 Figure 21. Representation of a confusion matrix with horizontal argument as Ground Truth (GT) and vertical as Prediction, showing correctly identified regions as TP, incorrectly identified regions as FP, missed regions as FN and correctly identified negative regions as TN >>>24

RkJQdWJsaXNoZXIy OTc5MDU=