"Advances in cancer imaging and technology"—special collection —introductory Editorial

Emerging technologies in oncological imaging: beyond morphology

BJR|Open is an open access international research journal published by the British Institute of Radiology. It is a multidisciplinary journal covering clinical and technical aspects of radiology, radiotherapy, radiation oncology, radiobiology and medical physics. We are delighted to announce an upcoming special collection dedicated to the advances in cancer imaging and technology, to be published in early 2023. This special collection invites submissions of original research, reviews, short communications and commentaries, and this introductory Editorial by our Guest Editors (Figure 1) will outline the aims and scope of the collection.

Figure 1.

From left to right: Professor Regina Beets-Tan, Professor Haibin Shi, Dr Zuhir Elkarghali and Dr Sharyn Katz

Radiological images are pivotal for screening, diagnosis, staging, and follow-up/monitoring in the oncological clinical workflow. ^1,2 Medical imaging has become more important in oncologic management with technological advances in the quality of imaging now possible and the increase in patient access to cross-sectional imaging. For example, imaging-based population screening programs have highlighted the importance of radiology in early-stage cancer detection. The reliance on medical imaging in oncology has been further potentiated by the paradigm shift in cancer care brought about by personalised medicine, which focuses on the individualised approach to oncologic management. With this, the number of requested scans has dramatically increased across all diseases. ^3,4 In the face of this increasing demand for imaging and expectations for the high precision in imaging assessments needed for personalised medicine, novel imaging techniques that can increase the speed of interpretation and yield of data gleaned from imaging have become a major focus for the radiological community.

Radiology is arguably the medical discipline most reliant on progress in computer technology. Digitalisation in radiology occurred decades before any other medical department. Subsequently, the infrastructure for experimenting and implementing new computer technologies is most mature in the domain of medical imaging. Emerging technological developments are being tested and validated largely on the back of existing radiological workflows. The barrier to implementing these novel technologies is often not technical but rather reluctance of adoption from clinical staff, who question the added value of these new developments.

A useful way to view these new technological trends is to reflect on past historical developments. Radiological images have come a long way from the first shadows on film that had to be processed in a developer. Now, radiologists can scroll through a three-dimensional volume of patient anatomy on a computer with slice thicknesses under 1 mm. The jump from two-dimensional images (e.g. X-rays) to three-dimensional volumes (e.g. CT and MRI) unlocked valuable spatial and morphological information to augment clinical decision-making.

Similarly, most new techniques aim to harness imaging to gain an even deeper insight into what is being imaged, which is essential to personalised medicine. Whether it is radiogenomics, cell tracking with X-nuclei MRI, or novel quantitative imaging modalities, most new techniques attempt to leverage the non-invasive nature of imaging to gain insight into the underlying biology/oncogenic process. Essentially, we might consider that the field of radiology is attempting to unlock a new “dimension” in imaging, where imaging no longer reflects just shapes but also biological processes. This development is urgently needed as we strive to harness imaging to identify imaging biomarkers that can predict and diagnose biological processes.

In this Editorial, we highlight several technological developments that have emerged in this field, and these topics are some of the areas we would like to encourage submissions to this special collection:

Radiomics: morphology as a biomarker

Radiomics is the domain of research where quantitative information is mined from routine clinical images to explore associations between morphological phenotypes and clinically relevant outcomes. ⁵ Radiomic features encode the morphology of the area of interest. In oncology, that is typically the tumour and peritumoral tissue. The full collection of radiomic features derived from a tumour is a kind of morphological phenotype. ⁶

The patterns of radiomic features extracted from these images reflect the tumour histologic structure and pathophysiology as well as its interaction with the host immune system, such as the presence or absence of tumour-infiltrating lymphocytes. This is particularly useful in the error of targeted therapy as we seek to stratify patients into groups most likely to benefit from targeting a specific metabolic pathway or functional interaction. In addition, radiomic analysis allows for non-invasive interrogation of the tumour ultrastructure in situ, which mitigates the shortcoming of sampling error from biopsies and allows for the study of the tumour features before it is disrupted by therapeutic intervention. In this way, radiomic signatures have been discerned that associate with prognosis, treatment response, and adverse event prediction for multiple therapies and tumour histologies. ^5,7

The emerging subdomain of radiomics termed “radiogenomics” has gained interest due to its potential complementary role to biopsy-based approaches. The driving hypothesis of radiogenomics is that tumour biology drives morphology, and morphological differences, as quantified by radiomics, can be used to differentiate between tumours of different biologies. It promises to overcome the limitations of tissue biopsy specimens by unlocking information from the entire tumour burden while still in situ allowing for the evaluation of tumour heterogeneity not only at diagnosis but longitudinally as new mutations emerge over the course of therapy. ⁸

Fully automated AI pipelines

While the work in radiomics highlights associations between morphology and (clinical) outcomes, the domain often cites automation and minimal human input, and thereby objectivity, as an advantage. However, the reality is that for most radiomics research, datasets are curated by humans, radiological images are manually segmented, features are pre-engineered/handcrafted, and the software pipelines are tailored to the project. In essence, “invisible hands” exert influence (and potentially bias) on the workflow.

With this drawback in mind, computer scientists have increasingly gravitated more towards end-to-end solutions (such as deep learning) where many of the steps of the radiomics pipeline (e.g. feature extraction) can occur automatically.

An example is the Prognostic AI Monitor (PAM) algorithm, which aims to predict prognosis under a given treatment. ^9,10 With traditional radiomic approaches, such a goal would be achieved by segmenting the tumour, extracting features, and modelling the features to the prognosis. Not only is there significant human intervention in the pipeline, but also the model is limited to learning from the tumour and ignoring the patient as a whole entity.

PAM works on the principle of needing as little human input as possible. Essentially, the algorithm takes serial images from a patient as input and generates deformation maps reflecting the global morphological changes that occurred in a patient, not just the tumour, during treatment. The model then uses these automatically generated deformation maps to predict prognosis. While still an algorithm in its infancy, PAM highlights the possibility of using routine imaging data without human intervention. As the field of medical AI matures, we expect more AI algorithms to follow the path of end-to-end prediction.

Photon counting CT: a novel CT technology

Photon counting CT is an emerging new imaging technology with the first scanners becoming available for clinical use. One major way that photon counting CT differs from conventional CT technology is that the incident photon striking the detector is converted directly into an electrical pulse, the height of which is proportional to the energy of the photon. By comparison, in conventional CT, the incident photon is first converted to visible light which then is transduced into an electrical signal. In addition, photon counters can set the threshold of detection above that of electronic noise and sort the energy of incident photons into energy bins according to the photon’s energy. This differs from conventional CT that measures and integrates all the photons striking during the measurement interval including photons of different energies and noise. A major benefit of this technology is the potential for substantially decreased image noise which is particularly useful for low-dose applications such as lung cancer screening and calcium-scoring CT.

Another major advantage of photon counter CT scanners is that, unlike conventional CT scanners which are limited in resolution by the minimum size of the manufacturability of increasingly smaller scintillator detector elements and septa, photon counters do not have separate detector elements making them easier to manufacturer. As a result, the spatial resolution limit for photon counters is an image spatial resolution limit of 0.07 × 0.07 mm² to 0.28 × 0.28 mm² compared to a lower limit of approximately 0.5 × 0.5 mm² for conventional scanners. ^11,12 This increased spatial resolution could impact the sensitivity of detection of small lesions, including small nodules and coronary plaques, and potentiate technologies such as radiomics which relies on the spatial resolution of the scanner to extract imaging features that reflect tissue structure. In addition, increased spatial resolution may be of value in temporal bone CT and the characterisation of indeterminate lesions.

Another useful feature of photon counting is the ability to bin photons by pre-set categories, such as the energies of water, iodine and calcium, and create a colour-encoded image of the relative tissue composition. This technique is known as material decomposition and allows for creating virtual non-contrast and material-specific overlays without the beam hardening effects observed with these techniques on conventional scanners. This can be of great value for a number of applications including breast CT and detection of tissue enhancement in indeterminate lesions.

In vivo multimodality imaging of disease

With the rapid development of medical imaging technology, molecular imaging has gradually emerged as a powerful means for visualising cellular function and uncovering the mechanism of physiological and pathological processes such as cancer, cerebral disease, and cardiovascular disease. ^13,14 Different types of imaging technologies including CT, ultrasound, positron emission tomography (PET), single photon emission CT (SPECT), are often applied in the course of disease diagnosis and treatment. ¹⁵ Nevertheless, each imaging modality has its own advantage and limitation. ^16–18 For example, MRI has high tissue contrast and spatial resolution, though it can be limited in sensitivity and specificity, in certain applications since the data it provides are restricted to the magnetic properties of the tissues and their morphology. ¹⁹ A molecular imaging modality such as SPECT or PET can monitor the distribution of radioactive tracers in a living system, providing information on the dynamics of tumour biology in situ, often with high sensitivity and specificity. ²⁰

One of the limitations of a molecular imaging modality such as SPECT or PET is that, because it measures the dynamics of a living system, it is susceptible to variables impacting tumour metabolism that can be difficult to control for, including medications, comorbidities, and effects of therapy. ²¹

However, while usually highly sensitive, the low spatial resolution of SPECT or PET can be limiting. One solution to this barrier is to acquire these modalities together with CT or sometimes MRI. ²² Multimodality imaging that integrates multiple imaging technologies can often provide increased accuracy through cross-validation across the modalities, each with different strengths and weaknesses. For example, highly FDG avid tissue may be revealed to be normal muscle tissue on CT. The specificity of molecular imaging can also be improved by careful selection of the molecular contrast agent (also called molecular imaging probe). The choice of molecular imaging probe used to selectively enhance the contrast between targeted tissues and the surrounding tissue is highly significant for sensitive and specific detection of the pathological region. ^23,24 Therefore, it is very meaningful to develop advanced molecular imaging probes with multimodality imaging and theranostics capability for early diagnosis and effective treatment of various diseases in vivo.

¹⁹F MRI: non-radioactive labelling and imaging

¹H MRI imaging is the modality of choice to examine soft tissue with high spatial resolution and tissue contrast, without radiation and is widely used in clinical imaging. Beyond the ¹H atom, there are other nuclei that can be detected via magnetic resonance. These so-called X-nuclei are a topic of interest in biomedical research because the information they provide goes beyond anatomy to give tissue specific molecular data. Physiologically relevant atoms such as sodium (²³Na), potassium (³⁹K) and chlorine (³⁵Cl) can be studied to uncover disease-specific alterations with diagnostic and prognostic potential. Unfortunately, non-proton nuclei are neither as sensitive to nuclear magnetic resonance nor as abundantly present as ¹H is in the body, making X-nuclei MRI technically challenging. However, developments in hardware (e.g. high-field MRI) and imaging methods have improved the quality of X-nuclei images.

¹⁹F MRI works on a similar principle to traditional proton MRI, but instead, it is the signal of excited fluorine-19 nuclei that are being detected. ¹⁹F MRI has the advantage of working on a virtually non-existent background signal level in the human body since free fluorine atoms do not exist in the body. Fluorine in the body is bound to the matrix of bone and teeth and hence is undetectable with conventional MRI. Introducing fluorinated agents will yield signals specific to that agent (and its target).

¹⁹F MRI can be used to visualise any targeted structure or cell in a background-free setting, as seen in endothelial cells. ²⁵ stem cells ²⁶ and immune cells. ^27–29 This additional layer of signal independent of ¹H omits the need for pre- and post-contrast images as required by superparamagnetic iron oxide nanoparticles (SPIONs) and gadolinium-based contrast agents. In addition, ¹⁹F is very sensitive to changes in the microenvironment which has been exploited to sense local changes in pH and pO₂ or detect enzymatic reactions. ³⁰ This great versatility makes ¹⁹F MRI an exciting field to explore in the era of personalised medicine.

¹⁹F MRI research is still ongoing and not yet ready for translation to routine patient care; however, exploratory studies of ¹⁹F MRI acquisition in humans have begun to emerge in the imaging literature ^31,32 and ¹⁹F contrast agents are now routinely manufactured with clinical-grade GMP standards. Strategies to improve signal-to-noise ratio and reduce imaging time through enhanced ¹⁹F probes, ³³ efficient pulse sequences, ³⁴ and compressed sensing are currently under development. ^35,36

Multispectral imaging is an exciting application of ¹⁹F MRI in relation to the tumour microenvironment. ^37,38 Here, it is possible to image different components or cell populations by labelling them with different “colours”, ^38–40 a functionality not currently achievable via PET. Multicolour ¹⁹F MRI relies on the wide chemical shift range of ¹⁹F agents, meaning that they possess different resonance frequencies which can be imaged separately. Combining distinctive ¹⁹F agents with select targeting ligands provides a much sought-after opportunity to examine the dynamic micro-environment non-invasively.

To conclude, this special collection aims to collate articles that cover a broad scope and content that may be relevant to the journal’s multidisciplinary audience, including radiologists, nucleair physicians, imaging researchers, physicists, radiographers, radiobiologists, and radiation oncologists. We encourage submissions of various article types covering topical areas in this field of cancer imaging and technology. We look forward to this special collection highlighting the advances in cancer imaging and technology.