Introduction

Uterine body cancers (UBC) are represented by endometrial carcinoma (EC) and uterine sarcoma (USa). EC are the most common female reproductive cancer in high-income countries with a growing incidence recorded in recent years [1]. USa are rare tumors and one of the deadliest gynecological cancers [2].

Clinical management of UBCs is complex and requires an accurate assessment of risk stratification factors in order to plan the correct therapeutic strategy. In UBCs the definition of risk categories is complicated by the stratification of tumors in different stages, different differentiation grades, different histological subtypes, and for EC, different biological subtypes based on molecular characteristics. Unfortunately, the evaluation of most of these parameters is operator-dependent, and therefore subject to possible inaccuracies. Furthermore, the need of introducing several parameters in the risk assessment, each of them associated with a certain risk of error, multiplies the probability of incorrect prognostic stratification.

For example, according to the European Society of Medical Oncology (ESMO), the main prognostic factors of EC are represented by disease stage, histological type, degree of differentiation, and vascular lymphatic invasion. However, the assessment of all these parameters is operator-dependent and poorly reproducible, even among experienced operators. All this leads to a non-negligible risk of incorrect classification of EC patients in relapse and death risk classes.

Moreover, additional risk stratification methods have become available over the last few years, based on the molecular features of TCGA [3, 4]. The TGCA published the first comprehensive genomic characterization of EC in 2013 and identified four subgroups: 1) ultramuted–POLEmut EC, harboring pathogenic mutations in the POLE gene, 2) hypermuted—Mismatch repair deficient (MMRd) EC showing a mismatch repair (MMR) defect or microsatellite instability (MSI), 3) copy-number low—mutated TP35 (p53abn) EC with mutations in TP53 and 4) copy-number high, a group with no specific molecular profile (NMSP).

However, this additional stratification system, while able to provide useful prognostic information, complicates the possibility of assessing the risk for the individual patient in daily clinical practice. In fact, an integrated assessment of traditional risk factors (stage, grading, lymphovascular infiltration) together with genetic risk factors is particularly complex and does not find clear indications from international guidelines. In this context, the introduction of predictive models able to manage all these data thanks to the use of AI methods, would be particularly useful.

Regarding USa, a common issue is the difficult differential diagnosis between USa and its benign counterpart (fibroid) [5]. Pathology examination, especially on the surgical specimen, is frequently the only method allowing a definitive clarification of diagnostic doubts. However, inappropriate resection techniques, such as intraoperative fragmentation of the tumor (morcellation), can lead to dramatically worse outcomes [6].

To solve these issues, standardized evaluation by AI algorithms able to overcome the human cognitive possibilities seems very attractive [7]. In this regard, radiomics and more generally AI-based analyses represent emerging translational research fields aimed at data mining from images and management of complex datasets to develop predictive models, respectively. Both of them, eventually combined with or including gene expression data, could support evidence-based diagnostic and therapeutic decision, in order to allow personalized and precision medicine in this setting [8].

Based on this background, the purpose of this protocol for systematic review is to provide an overview of radiomics and AI studies on UBCs (EC and USa).

Methods

Selection of studies

After duplicate publications removal, three authors will independently screen the retrieved papers based on their title/abstract/keywords to perform a preliminary selection based on the afore mentioned eligibility criteria. After this first screening, the remaining papers will be evaluated by examining the text in full. All differences arising at this stage will be solved by consensus between the researchers. In case of disagreement a fourth author will be involved in the final decision. Then, from the definitively selected papers data will be collected using a specific form including study design, number of patients, imaging type, AI or radiomics method, study objective, tumor histology and stage, validation group, segmentation type, previous therapies, clinical response evaluation criteria, and other prognostic factors. For the excluded papers, the reason for their removal from the analysis will be reported in the PRISMA flow chart (Fig 1). The final information will be verified by a lead author.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Shows the studies selection process.

https://doi.org/10.1371/journal.pone.0267727.g001

Discussion

The possibility of non-invasively extract biological and molecular information through images, using radiomics techniques, is obviously of great interest. In fact, the introduction of radiomics analyses in oncology shows growing interest and applications. The radiomics process can be divided into distinct steps such as image acquisition and reconstruction, image segmentation, features extraction and qualification, analysis, and model building [8]. Quantitative image features based on intensity, shape, size or volume, and texture offer information on tumor phenotype and microenvironment (or habitat) that is distinct from that provided by clinical reports, laboratory test results, and genomic or proteomic assays [13]. Furthermore, and more generally, AI is becoming a major player in integrating clinical, pathological, biological, molecular, genetic, and imaging data to develop predictive models [14].

The latter, requiring the use of machine learning or deep learning algorithms [15, 16], will be critical for oncological decision-making in a context where the amount of data available for each patient will be far beyond the human cognitive capacities.

In particular, AI-based analyses of radiological, histological and molecular features could improve the diagnostic and therapeutic pathway of UBCs. In fact, the availability of diagnostic and prognostic biomarkers is still an unmet clinical need in EC and USa [17, 18]. More specifically, integration of multi-omic data through AI systems could bypass the conventional clinical and molecular risk classifications (ESMO risk [19] and TCGA subclasses [3]) to allow a more personalized management of EC. In any case, even the simple use of traditional risk stratification systems requires radiological and pathological evaluations, both of which are subject to possible inaccuracies [20]. The latter can affect the reliability of tumor staging and prognosis prediction, with potential negative effects on clinical outcomes and possible legal disputes. In this context, the use of AI based systems has the theoretical potential to improve the reliability of individual diagnostic methods. For example, radiomics could be the key to solve the complex problem of differential diagnosis between uterine fibroid and USa.

In this scenario, a detailed review of available literature data is needed to summarize and analyze methods and results of AI applications, including radiomics, in the UBCs setting. In particular, the reliability of predictive models based on AI/radiomics will be evaluated based on the presence and type of patient populations used as validation cohorts (internal, external or prospective). Indeed, this point is critical in order to introduce these models in the clinical management of UBCs.

In conclusion, an increasing demand for precision and personalized therapies is required in oncology and particularly in the UBC setting. The aim of this protocol for systematic review is to evaluate all available information about applications of radiomics and AI-based models to improve diagnosis, staging, risk stratification, therapy and follow-up of UBCs.

Supporting information