Philippe Lambin

Dpt of Precision Medicine, Maastricht University, The Netherlands

In November 8, 1895, the German scientist named Wilhelm Röntgen, discovered an unknown type of radiation that he named ‘X-rays’. In fact, X-rays are still called ‘Röntgen Radiation’ in several languages, for example in German, Russian and Japanese. Röntgen was awarded the first Nobel Prize in Physics 1901 for his discovery which created the disciplines of radiology and nuclear medicine and paved the way for today’s Precision Medicine. Following the first famous photo of his hand’s wife, over a century of extensive use of X-rays have saved millions of lives across the globe. However, interpretation of radiological images has relied on the human eye. It took me years to understand that our eyes do not fully see the reality. Our sense of sight is not quantitative and what we “see” is typically simplified and contextualized by our visual cortex.

The idea of Radiomics was born in my lab in 2008 in Maastricht after several formal and informal discussions with Bob Gillies during his sabbatical as TEFAF Chair Recipient. There is an expression in Dutch that says, “Meten is weten” or “Measuring is knowing”. This is the foundation of Radiomics. Therefore, Radiomics, is synonymous with quantitative imaging: any imaging-based quantitative feature will do e.g. tumour volume, semantic features and deep features. In addition, in contrast to biopsy, Radiomics is based on three- and sometimes even four-dimensional information, when using a delta-Radiomics approach. It also has the benefit of being inexpensive and is easy to implement, compared to genomics and proteomics, because it uses existing images.

A few years ago, some experts predicted that handcrafted Radiomics, initially described in our 2012 (Eur J Cancer), 2014 (Nat Commun) & 2017 (Nat Rev Clin Oncol) papers, would disappear and be overtaken by ‘Deep Learning’, another type of quantitative imaging Radiomics approach. However, this did not happen mainly because Deep Learning is very data hungry and lacks the transparency of handcrafted Radiomics. Today, it is apparent that the medical community is not ready to prescribe or reimburse adjuvant chemotherapy or immunotherapy without understanding “why”. So the next challenge will be to combine handcrafted Radiomics and Deep Radiomics to make Radiomics more accurate (e.g.* https://doi.org/10.1038/s41467-022-30841-3 and https://doi.org/10.3389/fmed.2022.915243), explainable (the concept of “XAI”) and correlate it to biology. Another challenge will be to deal with the large heterogeneity of images, produced by different generations of hardware’s and different reconstruction algorithms.

I believe that the first clinical application of Radiomics will be for automatic segmentation and in answering diagnostic questions rather than predicting the future. I also believe that Radiomics will be part of the drug development process.

Radiomics is also a superb example of “Convergence Sciences” a combination of clinical questions, biological issues and technological solutions. (…)

Reference: Extract of the preface of the book 《FOUNDATIONS of RADIOMICS》2020 edited by Dr. Jie Tian, Director of the Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences.

*Add in 2022