Radiological imaging is essential for surgical oncologists: it is crucial to select candidates, define resectability and plan a safe and oncologically adequate intervention. Morphological characteristics of the tumors (i.e., number, size, and relationship with vessels) are well-known prognosticators closely associated with pathology data and form the basis for current patient-tailored treatments. However, modern oncology aims for precise, personalized care based on a comprehensive understanding of tumor biology. From this perspective, traditional imaging could seem outdated.

Over the past 10 years, radiomics has emerged as an innovative discipline within the field of oncological imaging, potentially re-assigning a key role to the radiologist. The growing interest in this topic has produced a remarkable annual publication growth rate of 120% [1]. Radiomics harnesses advanced computational algorithms and quantitative analysis techniques to extract features from manually or automatically segmented images obtained through various imaging modalities (US, CT, MRI or PET) [2]. In practice, radiomics treats images as data, and translates the radiological texture of the tissues into numbers. The underlying assumption of radiomics is that medical images can convey genomic and phenotypic information using a non-invasive approach. Therefore, the resulting information could drive oncological and surgical decisions. In the present Editorial, we will consider radiomics applied to liver diseases as paradigmatic of its potential clinical application (Fig. 1).

Fig. 1
figure 1

Current and future perspectives in the application of radiomics in liver surgery

Full size image

From consolidated evidence…

Even if most surgeons still consider radiomics a statistical trick rather than a clinically useful method, it has an established role in literature, a role that is consistent through different imaging modalities and liver diseases, with realistic potential clinical applications [3,4,5,6]. Firstly, radiomic analyses can differentiate benign and neoplastic hepatic lesions (e.g., lithiasis-related inflammatory mass vs. cholangiocarcinoma, and adenoma vs. focal nodular hyperplasia), and distinguish different tumors (i.e., HCC, cholangiocarcinoma, and mixed forms). Secondly, radiomics can produce correlates to pathological and genetic data in a non-invasive evaluation of the tumor biology, providing in the preoperative setting information that is usually available only after surgery. Such data are the determinants of prognosis, and, in a precision medicine perspective, they are the key to define a patient-tailored treatment strategy. Furthermore, radiomic indices per se may have a tight association with patient prognosis, becoming relevant biomarkers themselves. Thirdly, radiomics can prognosticate the response to systemic therapies after a few cycles or even before administration, much earlier than the standard RECIST criteria. It could drive the choice among different available treatment options and prevent/minimize toxicity due to ineffective treatment. Finally, radiomics may predict the risk of liver dysfunction in cirrhotic patients undergoing surgery thanks to its tight association with tissue characteristics. One further concept deserves attention. In many studies, radiomic analyses are complementary to—and do not replace—the clinical data. Often the most promising predictive models combine both clinical and radiomic features. Such results confirm that radiomics may convey new information that is not present in standard clinical evaluation.

… to new perspectives

Beside the aforementioned evidence, several additional paths of research are under investigation. Radiomics has the potential to capture both the intra-tumoral heterogeneity [7], unveiling resistance and progression mechanisms, and the inter-tumoral one, collecting and combining data through all tumor sites [8]. The radiomic analysis of the peritumoral tissue is a further breakthrough. Several biomarkers are present at the tumor/liver interface (immune infiltrate, tumor margin profile, satellite nodules, etc.). Preliminary data demonstrated that the peritumoral rim has a radiological pattern equal to the non-tumoral parenchyma but a specific radiomic profile, intermediate between the tumor and liver, which contributes to the prediction of both pathology data and survival in liver tumors [9, 10]. The analysis of the peritumoral tissue could contribute to solve the long-standing debate among surgeons about the adequate width of the surgical margin to prevent any local recurrence risk. Recently, some authors even considered the radiomic analysis of the whole non-tumoral liver to predict the onset of liver metastases or tumor recurrence when no nodule is evident yet [11]. Finally, in the current issue of Updates in Surgery, we had the privilege to advance a further proposal. The hepatologists routinely perform a biopsy of the liver parenchyma to study its characteristics. Why not adopt a virtual liver biopsy, i.e., the radiomic analysis of a standardized volume of non-tumoral liver parenchyma? We demonstrated that such analysis can predict the characteristics of the liver and the risk of liver dysfunction and bile leak after elective hepatectomy [12].

Back to reality

Despite the available encouraging data, radiomics-based tools are not yet part of our clinical practice. How to fill this gap? We hypothesize a three-step process based on some important requirements that should be satisfied. Firstly, radiomics should meaningfully improve patient outcomes or, at least, have a major impact on clinical processes [13]. Secondly, the radiomics tool should be standardized and needs to be certified for clinical use. Thirdly, the radiomics tool should be easy enough to be integrated into the clinical routine.

The first requirement is fulfilled by a high-quality demonstration of efficacy [14], e.g., in a randomized clinical trial (RCT). Because RCTs are expensive to set up and conduct, only the most promising tools will enter a trial. These tools should be identified in external validation, which is comparatively cheap. Nevertheless, validation on new datasets is not always performed by the authors themselves. This is the point where many radiomics tools fail translation, for two main reasons. First, the tool may not generalize to new datasets because it lacks robustness against heterogeneity in imaging devices, protocols, expert segmentation, radiomics software, etc. Its claimed performance can therefore not be achieved in external settings. Secondly, studies may not be reported in sufficient detail to allow replication of the analyses [15].

The second and third requirements for clinical translation (standardization and usability) are easier to fulfill, and part of the commercialization process. Twenty-five research teams adhered to the IBSI (Image Biomarker Standardisation Initiative) and provided a standardized and reproducible computation of 169 radiomic indices [16]. Efforts are ongoing to harmonize data from different image acquisitions. Furthermore, important steps of the advanced imaging and analysis can be automated. For example, radiomics software are now integrated in a picture archiving and communication system [17].

From PACS to beside: the near future

Though the aforementioned clinical gap remains to be closed, the consistent positive results demonstrate an undeniable clinical role for radiomics. Current advances in semi-automatic and automatic segmentation will help accelerate and harmonize the data collection. Further, ongoing digitization efforts and implementation of IT platforms to support AI applications will ease the inclusion of radiomics-based prognostic models and decisional protocols. New perspectives could even refine the impact of radiomics. The dynamic evaluation of tumors through multiple images over time or through different imaging modalities could enhance radiomic predictions. Radiomic maps [7] pave the way to the translation of statistical indices in interpretable images, and, consequently, in more intuitive and usable biomarkers. In the near future, the multidisciplinary team for liver tumors and the HPB surgeons will require the expertise of radiological oncologists. These specialists will not only possess proficiency in the morphological evaluation of the tumors but also in their biological assessment, incorporating radiomic indices as part of the standard biomarker evaluation process.