In the world of rapidly growing diseases affecting the human body, the risk factors of cardiovascular diseases are increasing at an alarming and higher rate than most other diseases. Hence, automatic and fast detection of heart diseases is important for better treatment. This paper presents a methodology to detect and classify different cardiovascular diseases using deep neural network. All of the 12-lead ECG signals are first combined to form a complete ECG signal. Then the ECG signal is decomposed into different modes using the Variational Mode Decomposition (VMD) method. The first three higher-order modes are combined to form a modified signal, which is named as a modified ECG signal. It is estimated that higher-order modes contain more significant information than other modes. Next, the modified ECG signal is fed into a one-dimensional CNN network to learn the features of the signal. Finally, the network classifies the signal into different disease domains through its softmax regressor at the outlet of it. It seems that the one-dimensional CNN learns the features of the modified ECG better than those of the original signal. The method is applied to the Physionet CINC challenge-2020 dataset. The class-weighted F_2 score of the method on the test dataset is 0.831 and the Jaccard-measure (G_2 score) is 0.623.