Our findings were noteworthy, particularly in the realm of model performance and efficiency. The active learning methodology, initiated by query 2 in our training set, significantly enhanced the performance of our segmentation model. Notably, this improvement was not just in accuracy but also in the reduction of preprocessing time required for model creation, signifying a leap in efficiency.

In the testing set, our model demonstrated exceptional performance with an Area Under the Receiver Operating Characteristic (AUROC) of 0.99 for classification using both the ResNet and EfficientNet frameworks. For anomaly detection, the model achieved an AUROC of 0.88 with ResNet and 0.91 with EfficientNet. In terms of the Area Under the Precision-Recall Curve (AUPRC), the values were equally impressive: 0.99 for classification (ResNet and EfficientNet), 0.83 for anomaly detection (ResNet), and 0.77 for anomaly detection (EfficientNet).