Editorial: The Application of Radiomics and Artificial Intelligence in Cancer Imaging

Cancer provides a unique medical decision context considering its variegated forms with the evolution of the disease, as well as the individual condition of patients, their ability to receive treatment, and their responses to treatment. Technological advances in medical imaging bring benefits to address the challenges of accurate detection, characterization, and monitoring of cancer, but traditional imaging assessment of cancer commonly relies on visual evaluations, the interpretations of which may be augmented by advanced computational analyses. Radiomics and artificial intelligence (AI) promises to make great strides in the qualitative and quantitative interpretation of cancer imaging. Radiomics, the high-throughput mining of quantitative image features from standard-of-care medical imaging, is gaining significant importance in cancer research, while AI excels at distinguishing complex patterns in cancer images and thus provides the opportunity to alter image interpretation from a purely qualitative and subjective task to one that can be quantified and effortlessly reproduced. Therefore, this Research Topic recruited studies that explore the application of radiomics and AI in cancer imaging. We are so glad to see that many wonderful works were submitted to our Research Topic. In the end, a total of 45 papers were published, and all papers were original studies. The studies were conducted in various countries, including China, the USA, Italy, Australia, Spain, Germany, and the Netherlands. The authors explored different methods to explore the role of radiomics and AI in cancer imaging.


INTRODUCTION
Cancer provides a unique medical decision context considering its variegated forms with the evolution of the disease, as well as the individual condition of patients, their ability to receive treatment, and their responses to treatment. Technological advances in medical imaging bring benefits to address the challenges of accurate detection, characterization, and monitoring of cancer, but traditional imaging assessment of cancer commonly relies on visual evaluations, the interpretations of which may be augmented by advanced computational analyses. Radiomics and artificial intelligence (AI) promises to make great strides in the qualitative and quantitative interpretation of cancer imaging. Radiomics, the high-throughput mining of quantitative image features from standard-of-care medical imaging, is gaining significant importance in cancer research, while AI excels at distinguishing complex patterns in cancer images and thus provides the opportunity to alter image interpretation from a purely qualitative and subjective task to one that can be quantified and effortlessly reproduced. Therefore, this Research Topic recruited studies that explore the application of radiomics and AI in cancer imaging.
We are so glad to see that many wonderful works were submitted to our Research Topic. In the end, a total of 45 papers were published, and all papers were original studies. The studies were conducted in various countries, including China, the USA, Italy, Australia, Spain, Germany, and the Netherlands. The authors explored different methods to explore the role of radiomics and AI in cancer imaging.

Studies With Radiomics in Cancer Imaging
Radiomics aims to extract mineable high-dimensional imaging features from medical images, and it enables data to be extracted and applied within clinical-decision support systems to improve the accuracy of cancer diagnosis, prognosis, and prediction (Shan et al.; Shi et al.; Gao et al.). This type of studies included in the Research Topic used features extracted from different kinds of images, Positron Emission Tomography-Computed Tomography (PET/CT), CT, Magnetic Resonance Imaging (MRI), and Contrast-Enhanced Spectral Mammography, to enhance the performance of

CONCLUSION
In conclusion, radiomic features have the potential to obtain biological and pathophysiological information from ROIs, and the corresponding quantitative features can provide rapid and accurate non-invasive biomarkers for cancer diagnostics, prognosis, and treatment response monitoring. AI has got a lot of attention, especially, for the success of deep learning to create complex neural architectures to solve difficult problems which would be impossible with traditional machine learning methods, and AI-based methods have shown significant progress in the field of radiological-based medical imaging applications. However, the application of AI-based methods in cancer imaging to date has not been vigorously validated for reproducibility and generalizability, there are many challenges that still remain: 1) A large number of labeled images are needed to build generalizable robust AI models, while it is timeconsuming to annotate large-scale medical image datasets like ImageNet dataset. To tackle this issue, data augmentation can be used to increase the amount of data by warping, rotating, or inverting existing images, or creating synthetic data from existing date using Generative Adversarial Networks (GANs). Another widely used approach is transfer learning, and it has been established as one of the most practical paradigms in medical image processing with insufficient training samples since it makes full use of the parameters of model pre-trained from large-scale natural image datasets. Due to the big inter-domain discrepancies between natural images and medical images, self-supervised learning provides one possible solution to overcome the limitations of transfer learning, and it learns representations in a self-supervised way which is beneficial to medical image processing. 2) The black-box nature of DL methods is one of the largest stumbling blocks to the wider acceptance of DL for clinical applications. Even when the DL-based method shows good performance in many cases, it is difficult or almost impossible to explain how the networks perform various tasks. In recent years, the attention mechanism is explored to interpret DL Models, which attempts to build weights that reflect which part of the input is more important for decision making. 3) In real-world clinical applications, there are other issues including ethical, regulatory, and legal issues to solve, which should be carefully considered for the development of AI models in cancer imaging.
This Research Topic involved many wonderful works, which made full use of radiomics and AI in cancer imaging. We appreciate all the reviewers and authors for their contributions to this Research Topic, and we hope this Research Topic can gain more attention in the related fields.