Case-based lung image categorization and retrieval for interstitial lung diseases: clinical workflows.

Purpose: Clinical workflows and user interfaces of image-based computer-aided diagnosis (CAD) for interstitial lung diseases in high-resolution computed tomography are introduced and discussed.

Methods: Three use cases are implemented to assist students, radiologists, and physicians in the diagnosis workup of interstitial lung diseases.

Results: In a first step, the proposed system shows a three-dimensional map of categorized lung tissue patterns with quantification of the diseases based on texture analysis of the lung parenchyma. Then, based on the proportions of abnormal and normal lung tissue as well as clinical data of the patients, retrieval of similar cases is enabled using a multimodal distance aggregating content-based image retrieval (CBIR) and text-based information search. The global system leads to a hybrid detection-CBIR-based CAD, where detection-based and CBIR-based CAD show to be complementary both on the user's side and on the algorithmic side.

Conclusions: The proposed approach is in accordance with the classical workflow of clinicians searching for similar cases in textbooks and personal collections. The developed system enables objective and customizable inter-case similarity assessment, and the performance measures obtained with a leave-one-patient-out cross-validation (LOPO CV) are representative of a clinical usage of the system.