What is the best method of combining image feature and numeric feature together using CNN in Machine Learning?

Question

I've got this question here: For example, if it is necessary to predict a disease using both image data and some numeric data, so that the features would be like:

1. feature 1: image of the disease. in shape: (batch_size, width,height)
2. feature 2: numeric data about the patient(age,height, sex, country, salary...) in shape: (batch_size,number_of_numeric_features)

and the output of the model should be 0/1, 0 is healthy, 1 is sick.

I know one way is to use the flat feature as a shape: (width*height+number_of_numeric_feature) in this case the advantage of CNN in image classification won't be utilized. (a feedforward network)

So my question is: is there a best solution to combine image feature and numeric feature using CNN? Would adding numeric features as image pixels in one channel of the CNN feature helpful? in such case the positional distance of the numeric feature as pixels won't make any sense since they don't have relationship in distance of two pixels.

Answer 1

You should not use SUCH numerical data with CNN, as you mentioned yourself, it won't make any sense, but there is a way in which you could use your image with CNN, and use another network (e.g. MLP) for the numerical data, at the end, you can combine the output of MLP and CNN together and feed them to another MLP, or just take averages from their output and compare the results.