Over the past few years, multiple waves of COVID-19 (C19) have impacted millions globally. The optimal treatment for C19 remains uncertain, as vaccinated individuals have also experienced illness. C19 can be diagnosed swiftly and precisely, thereby preserving lives and shielding patients from expensive therapies. Researchers have used different types of clinical imaging, like chest CT scans (CCT) and X-rays (CXR), to find C-19. However, using ECG images to find people who are infected with C19 has gotten less attention. We utilize ECG samples for diagnosing C19 due to their accessibility, in contrast to CCT and CXR images. Researchers frequently transform ECG data into integer format prior to utilizing any methodology, leading to increased processing overhead. Accurately and swiftly identifying C19 using the ECG modality is both challenging and time-consuming. In this study, we used paper-based ECG images in an improved hybrid deep learning model called ResNet50+BiLSTM to try to solve these problems. We used BiLSTM to label ECG images as C19 with no findings after removing the last softmax layer and immobilizing the two residual connections after the first residual block in ResNet50. The ECG images are pre-processed using hexaxial feature mapping to convert paper-based EEG images into 2D colored images. We then use ResNet50 to effectively extract the features of the processed images, aiding in the classification process. We use LGBM for feature-based classification in this process. This combination provides dependable feature extraction with reduced training configurations. The Enhanced BiLSTM achieved the best accuracy of 97.22%, precision of 97.02%, sensitivity of 97.00%, and an F1 score of 96.84%. Ultimately, we employed heat maps to illustrate the images and understand the decision-making process of the Enhanced BiLSTM. This enhanced hybrid deep learning model assists physicians in delivering appropriate patient care.