_{Published January 4, 2021}

Clinical trials are an indispensable component in medicine. These experiments are usually aimed to assess or optimize the efficacy and safety of a particular medical procedure. The medical procedure could be a vaccine, a drug treatment, a diagnosis tool, etc. In the context of a diagnostic method, and particularly in medical imaging, clinical trial development and applications have been limited due to multiple factors. These trials are usually expensive and time-consuming. When a trial is designed for human subjects, there are various factors that need to be considered to minimize the risk to the patients, which limits the thoroughness of the study. Beside this, the ground truth or baseline state of the patient is often unknown, which significantly limits researchers from assessing technologies in an accurate and precise manner. Virtual imaging trials offer an alternative solution to overcome these limitations by providing a mechanism to conduct trials in silico, as summarized in a recent review paper, as well as an editorial that highlights an existential need for virtual imaging in medicine.

In this article, we briefly introduce virtual clinical trials in the context of medical imaging, highlighting some of the recent advancements and ongoing projects in the field.

What Are Virtual Imaging Trials?

Virtual imaging trials (VITs) involve conducting imaging research in silicousing mathematical models and computer simulations. This is done by mimicking the chain of clinical imaging trial processes, which includes patients, scanners, and interpretations. In a VIT, patients are defined with anthropomorphic, computational phantoms. These mathematically defined phantoms model the patient’s anatomy and physiology, typically based on clinical images and morphometrical and physiologic measurements. Another component in a VIT is the virtual imaging systems. They are designed to virtually image the computational phantoms, aiming to incorporate realistic models of the geometrical, physical, and processing attributes of different scanners. The last component of a VIT is the virtual interpretation in which a medical image is quantitatively assessed for a specific clinical task (e.g., lesion detection or disease quantification). Computer-based algorithms, set to mimic radiologist tasks, provide a means to interpret the simulated images. 

New Models for Disease

A major part of a VIT is the virtual patient cohort. Virtual patients have been modeled and advanced for several decades. To be relevant for specific clinical applications, virtual patients need to be representative in terms of anatomy, physiology, patient-to-patient variability, and pathology. As the current phantoms are getting more anatomically realistic and diverse, more efforts are being made to model targeted pathologies for specific imaging trial questions.

An evident example is the recently developed human models with coronavirus disease 2019 (COVID-19) pathology. These pathological models are based on clinical CT images of reverse transcription–polymerase chain reaction confirmed COVID-19 cases (Fig. 1). The shapes and density of the COVID-19 lesions are modeled based on the segmentation of real data. These lesions are incorporated into an existing virtual phantom library, creating a population of virtual patients with COVID-19 abnormalities. Such patient models can be used to design optimal imaging protocols for the diagnosis and staging of COVID-19, as well as other diseases.

• 

Automation and Components Integration

Since VIT components are modeled in a modular fashion, they need to be integrated with each other to conduct a complete VIT. This process is often burdensome and requires some manual involvement. Lack of integration often limits the potential of VITs to simulate a large number of patients and conditions. To circumvent this challenge, investigators are making the VIT developments as automated as possible. For instance, recent, fully automated segmentation algorithms enable researchers to generate patient-specific phantoms at large, without a need to manually or semi-manually model them. Beyond automation, each component needs to have standard inputs and outputs, facilitating their seamless integration with other components of a VIT.

Recent Applications

VIT toolsets are increasingly being used to investigate clinical imaging problems. In a recent study, the effects of beam collimation, pitch, and patient motion on the quality of CT images were quantified. The study was done by creating dynamic models of patients with cardiac and respiratory motions. The results showed that increasing beam collimation affects the Hounsfield unit accuracy, largely due to the presence of imaging scatter. In addition, morphology radiomics features were found to be largely unaffected by beam collimation (19–38 mm) and pitch values (0.5–1.5). Such studies cannot be done using patient images, as they would need multiple acquisition of the same subject at various imaging settings—not ethically attainable with real patients. They also require the full knowledge of patients’ dynamic anatomy and physiology, which is not known in real patients.

Another recent study utilized VIT tools to compare the COVID-19 mortality risk versus the radiation risk for chest CT or chest radiography imaging for different age cohorts. Since the pandemic started, the roles of x-ray imaging and its radiation risk to patients have been challenged. These comparisons seek to isolate the radiation-use justification of x-ray imaging in the diagnosis and treatment evaluation of the disease, irrespective of potential benefits. The COVID-19 mortality was extracted from epidemiological data. The radiation risk was estimated using a virtual imaging trial approach. In particular, organ doses were calculated using Monte Carlo simulations of a library of patient models. Patient-specific organ doses were then used to calculate risk index that was converted into an upper bound for related mortality rate, following National Cancer Institute Surveillance, Epidemiology, and End Results Program data. The study clearly shows that, by and large, particularly for older patients, the risk of COVID-19 far outweighs that of any potential radiation risk. These results are being supplemented by virtual COVID-19 imaging that includes image quality, thus enabling a complete assessment of risk and benefit of using imaging for the pandemic.

As medicine advances, patient care is evolving towards more targeted and patient-specific solutions. As such, imaging research needs to be adapted to account for patient specificity. A learning-based framework, known as iPhantom, has recently been developed to enable fully automated creation of a digital twin for a given patient. Given a computationally defined model of the patient, imaging procedures can be tailored individually to the patient, providing an optimal diagnosis with minimal radiation risk.

VITs have proven to be a promising tool that enable effective, robust, and objective imaging studies that would not be possible otherwise. Significantly faster and less costly, VITs can inform more targeted clinical imaging trials, making them more feasible and efficient. VITs can be used to inform the optimum design of state-of-the-art and emerging technologies. With all the benefits and potential applications, VITs need to be expanded and advanced even more to be applicable for different imaging modalities, patient cohorts, and applications. Given enough realism, VITs may even replace clinical trials in the future.