In my view, the main challenge is to retrieve lung images which are not just “visually similar” but rather have the same diagnosis.

 As for the image modality, I would not expect much utility of CBIR based on lung X-Ray whereas CT could be useful. In both occasions you need to segment images first, i.e., extract the lung component. No much sense to search for similar cases among the whole images what returns similar body constitutions but not similar lung disease.

 Remember also that CBIR is a potentially effective CAD tools ONLY under condition of VERY LARGE image database where really similar cases rather to occur. Few thousands of image items is a good amount to start with.

It depends what you are looking for. The specific goal of the CBIR for the lung (x-ray or CT) has to come from the clinical side. For instance, the challenge for finding similar cases when we are dealing with pulmonary nodules is mainly the low discrimination between the suspicious mass and surrounding regions, specially in x-ray. In such cases some sort of pre-processing is required first to amplify the nodule boundary against the surroundings. The actual search can be executed afterwards.    

Pls. see the work of A. Depeursinge, Sierre, Switzerland.

https://scholar.google.fr/citations?user=d6PRobAAAAAJ&hl=en

Best

Daniel

It can be interesting too ( https://lndb.grand-challenge.org/ )