RadEx: An open source python package for nonlinear radon transformation

Abstract

Effective feature extraction from medical images is important for improving disease detection and assessment. Conventional linear transforms, such as the Radon transform, may not fully capture subtle and complex nonlinear features present in medical imaging data. To address these limitations, we present RadEx, a nonlinear extension of the Radon transform. RadEx employs parameterized nonlinear projections to facilitate the extraction of additional nonlinear feature representations from imaging modalities such as chest X-rays and retinal fundus images. Initial evaluations indicate that RadEx can offer improvements over traditional Radon transforms and raw image-based approaches in disease classification tasks, including COVID-19 detection from chest X-rays and diabetic retinopathy grading from retinal images. By capturing more complex structural and nonlinear patterns, RadEx may support enhanced diagnostic performance and illustrates the potential benefit of integrating adaptive mathematical transformations into medical imaging workflows.