The result we got from running this network is 75 percent accurate on the validation set. This is not very good because of the criticality of the network usage. In medicine, there is not much room for error because a person's medical condition is on the line.
To make this accuracy better, we need to define a different criterion for evaluation. You can read more about it here:
https://en.wikipedia.org/wiki/Confusion_matrix
Also, you can balance the dataset. What we have now is an unbalanced dataset in which the number of diseased patients is much lower than the number of normal patients. Thus, the network becomes more sensitive to normal patients' features and less sensitive to diseased patients' features.
To fix this problem, we can SMOTE our dataset. SMOTing is basically replicating the data of less frequent classes (flipping the image horizontally...
