Myocardial ischemia (MI) is a heart disease that can cause various types of fatal heart attacks. Patients often miss the best treatment time when they develop the symptoms of a heart attack. Early detection of MI is considered to be necessary for curbing the deterioration of heart diseases because it is difficult to observe the symptoms of a heart attack through a medical check-up. Infrared thermal images of 165 healthy patients with different degrees of MI are collected, and all the samples are divided into training set and test set. Further, the geometrical differences between the left and right sides of the precordial area are extracted based on the geometric positioning of the infrared thermal image of a particular human body. Additionally, several convolutional kernels are used to reduce the dimensionality of the temperature difference set.The training set is trained using the back-propagation (BP) neural network based on the cross-validation method, and the network parameters are determined for establishing a classification model. After the 3×3 size Gaussian kernel operator is convolved on the temperature difference set, the classification accuracy of the test set with respect to the BP neural network becomes 95.56%, thereby denoting that the predictions for the new sample are considerably accurate. Further, the proposed method can rapidly and accurately assist during the early detection of MI in a clinical examination and provide a new methodology for the pre-diagnosis of MI.