Image class translation: visual inspection of class-specific hypotheticals and classification based on translation...

A major barrier to the implementation of artificial intelligence for medical applications is automated CNNs' lack of explainability and high confidence for incorrect decisions, specifically with out-of-domain samples. We propose a generalization of image translation networks for image classification and demonstrate translation networks' potential as a more interpretable alternative to conventional black-box classifiers.

We train an image-to-image network to translate an input image to class-specific hypotheticals, and then compare these with the input, both visually and quantitatively. Translation distances, the degree of alteration needed to conform to one class or another, are examined for clusters and trends, and used as a simple low-dimensional feature vector for classification.

On melanoma/benign dermoscopy images, a translation distance classifier achieved 80% accuracy using only a two-dimensional feature space (versus 85% for a conventional CNN using a $\sim 62,000$ -dimensional feature space). Visual inspection of rendered images revealed dataset biases, such as more scale bars in melanoma photographs than in benign lesions. Image distributions in translation distance space revealed a natural separation along the lines of dermatologist decision to biopsy, rather than between malignant and benign. On bone marrow cytology images, translation distance classifiers outperformed a conventional CNN in both 3-class (92% accuracy versus 89% for CNN) and 6-class (90% versus 86% for CNN) scenarios.

This proof-of-concept shows the potential for image-to-image translation to go beyond artistic/stylistic changes and to expose dataset biases, perform dimension reduction and dataset visualization, and in some cases, potentially outperform conventional end-to-end CNN classifiers.