Few-shot medical image classification with simple shape and texture text descriptors using vision-language models.