Automated Cardiomegaly Detection from Chest Xrays using Deep Convolutional Neural Networks: A Novel Approach for Improved Diagnosis

Cardiomegaly is an enlargement of the heart and is associated with many serious pointing to heart disease. In general, the defect requires early detection to avoid grave complications. This paper discusses a study in the application of CNN in the prediction of cardiomegaly using medical images. The CNN models are best for image-based diagnosis, since the features and patterns needed from the medical data are automatically extracted by the systems themselves. In this work, a binary classification methodology has been pursued to classify cases as either cardiomegaly (true) or no cardiomegaly (false). There were 1114 samples in the dataset, equally divided into true and false classes. Until this model eventually attained an overall accuracy of 80%, the precision and recall demonstrated relatively balanced performances in both classes. The precision on cardiomegaly cases was 0.78, while the recall was 0.83, with an F1-score of 0.80. On the other hand, non-cardiomegaly cases were found to have a precision of 0.82 and a recall of 0.77, with an F1-score of 0.79. The macro and weighted averages are both 0.80 for the precision, recall, and F1-score. These results also show the performance of CNNs in predicting cardiomegaly and, therefore, provide clinicians with a better tool to help diagnose heart enlargement. This suggests that the model has found a good balance between the two classes in that it was able to detect cases of cardiomegaly without allowing a large proportion of false positives or false negatives. This could potentially be improved even further by enhancing the model architecture and increasing the amount of data used for training to further enhance its accuracy in clinical usefulness.