Persistence landscapes could help achieve unbiased and accurate radiological interpretation

A new editorial was published in Oncotarget’s Volume 15 on November 12, 2024, entitled “Persistence landscapes: Charting a path to unbiased radiological interpretation.”

In this editorial, Yashbir Singh, Colleen Farrelly, Quincy A. Hathaway, and Gunnar Carlsson from the Department of Radiology, Mayo Clinic (Rochester, MN), introduce persistence landscapes, a mathematical method designed to address biases in medical imaging and artificial intelligence (AI). Persistence landscapes build on persistence images, which track how patterns in data appear and disappear across different scales. By transforming this complex data into simpler, more manageable forms, persistence landscapes create a format that is easy to analyze and compare. This makes it a valuable tool for identifying and correcting biases in medical imaging.

Medical imaging plays a critical role in healthcare, but it is not perfect. Biases, caused by differences in equipment, technology, or even the patient population, can lead to inaccurate diagnoses. Persistence landscapes offer a way to identify and fix these hidden issues. 

“[…] persistence landscapes have the potential to play a crucial role in identifying and mitigating biases in radiological practice, whether these biases stem from demographic factors, equipment variations, or the limitations of AI algorithms.”

Persistence landscapes are particularly effective at reducing random noise in medical images while preserving important details. This makes it easier for clinicians and researchers to focus on the most meaningful parts of an image. The method also improves AI tools by addressing common problems, such as when models are too focused on specific details or when they miss important information. Additionally, persistence landscapes also simplify the integration of data from different scan types, like positron emission tomography (PET) and magnetic resonance imaging (MRI), without introducing new errors.

Despite its potential, the use of persistence landscapes in real-world medical imaging comes with challenges. It requires powerful computers to process large data, which can be costly and time-consuming, and expert interpretation for meaningful use. Better tools are needed to make this method more accessible for clinicians. While integrating this method into clinical settings will take effort, the benefits could be transformative. With further research and refinement, persistence landscapes hold enormous promise for advancing equitable healthcare.

“Persistence landscapes represent a powerful new tool in our ongoing efforts to achieve unbiased and accurate radiological interpretation.”