The electrocardiogram (ECG) is an indispensable tool for diagnosing cardiovascular diseases (CVDs) because it can quickly and non-invasively monitor heart rhythms by recording electrical signals in the heart. Recent research has demonstrated that innovative machine learning and AI approaches have great potential to advance ECG interpretation to improve diagnostic capabilities. Despite these advances, more widespread application of such techniques is hindered by several factors. For instance, in many resource-limited areas, digital copies of ECGs remain unavailable. Additionally, ECGs scanned from paper realistically contain noise such as shadows or creases, which are often not accounted for in digital approaches. Furthermore, contemporary research is concentrated on detecting single conditions or diseases, despite the possibility of these co-occurring. To address these shortcomings, this work investigates the application of pretrained Convolutional Neural Networks (CNNs) by finetuning popular models such as AlexNet and ResNet separately for multi-label classification of realistic and noisy ECG images, aiming to enhance accessibility and diagnostic accuracy in real-world scenarios. This is achieved by leveraging ECG-Image-Kit to generate synthetic images from ECG images with various distortions. This research is based on The George B. Moody PhysioNet Challenge 2024 and received an F-score of 0.516 on its private validation set. This paper demonstrates that image-based ECG classification can identify multiple co-occurring cardiovascular conditions, offering a practical solution for areas where digital ECGs are not feasible. The results underscore the potential for the improvement of image-based models and suggest that future research should target this approach to provide a scalable method for interpreting ECG images in diverse healthcare settings. Our Team name is ECG_UVA and got the classification F-measure of 0.516 on the leaderboard on the validation dataset in the official phase.