DiaBeatNet-TCL: Diabetes Diagnosis Prediction of Pregnant Women from ECG Signal using CNN-LSTM Deep Learning Techniques

Diabetes that only occurs during pregnancy when blood sugar levels become too high is known as gestational diabetes. In low-contrast and noisy computed tomography (CT) scans, traditional networks frequently fail to accurately recognize tumor boundaries, resulting in inaccurate localization and classification. To overcome these limitations, a novel DiaBeatNet-TCL (Diabetes heartbeat signal using Time domain and CNN-LSTM) has been proposed for identifying diabetes in pregnant women in its early stages using ECG signals. The ECG signals are taken as the input, then the signals are pre-processed using the low-pass and high-pass filters to remove noise in the signal. The feature extraction is done in the Time domain, which is used to find the heart rate of pregnant women. A DL technique that uses CNN and LSTM combined to improve prediction accuracy. The CNN is used to extract the important patterns like QRS and T-wave, while LSTM is used as a memory to store the heartbeat intervals. This combined method efficiently classify the diabetics of the pregnancy women with diabetes. From the experimental results, the DiaBeatNet-TCL approach achieves an accuracy of 99.61% and an F1 Score of 93.23% for diabetes detection. The DiaBeatNet-TCL approach increases the overall accuracy of 9.27%, 6.72%, 4.52%, and 2.05% over LeNet, AlexNet, DenseNet and ResNet, respectively. The DiaBeatNet-TCL approach enhances the total accuracy of 5.63%, 0.37%, 4.85%, 5.96% better than SVM-CNN+LSTM, Attention R2W-Net, ECG-HbA1c, and ProWSy, respectively.