Hi Nitsa, that depends on how you are working with the DICOM data. If you want to train with the image data directly from files, this usually requires you to construct an ImageDataStore. For this, you would need to convert the DCM files to another format, like JPG. You can use the 'dicomread' function to read in the image from a DCM file, and then save it afterwards as another format.
However, if you are using the DICOM info and image in some other way, you can get that information with the 'dicominfo' and 'dicomread' functions, and structure your classifier however you like.