Medical imaging is on the threshold of a new era, thanks to new boundary-pushing tools such as artificial intelligence, machine learning and big data.
Accelerated processing speed is essential for creating high-quality images. Even as new imaging techniques provide greater anatomical and clinical details on the one hand, radiologists, oncologists and other diagnosticians also get faster access to imaging reports. This is one of the places where deep learning and artificial intelligence play crucial roles. These tools help to bring relevant information out of the electronic medical record and present it in a meaningful way, facilitating better-informed clinical judgment. Incorporating this information directly into the report can really add value as radiologists use deep learning. It will not only streamline workflows, but also be a major step towards more personalized medicine in radiology, experts in the field say.
Besides, incorporating AI and deep learning into operating systems helps to automate workflow. Automation is especially important for measurement-intensive procedures in specialties like cardiology and ob-gyn.
Diving deep down images
It was in late 2000 that the industry started taking notice of the high relevance and applicability of AI to medical imaging. The first system with embedded AI capabilities, the Logiq E10 by GE Healthcare ultrasound, secured approval from USFDA a decade later. This followed the CAD platform for evaluation of breast abnormalities, software meant for detection of diabetic retinopathy (IDx), AI triage software for stroke detection (Viz.AI), wrist fracture CADx software (Imagen Tech) etc. The USFDA decisions represented major drivers to global market development.
Since then, there has been an explosion of technologies that led to automation, acceleration, augmentation, on-demand access, intelligent machines and cognitive workflow applications.
Other than accelerating basic imaging exams, AI-based image analysis is employed in diverse scenarios such as high-volume routine imaging, time-sensitive imaging (especially in trauma cases) as well as for enhancing complex investigations.
CT, MRI and X-ray are the most preferred modalities for AI imaging companies. As many as 41 companies are tracking these types of images first, according to
the report Artificial Intelligence for Medical Image Analysis — Companies-to-Action, 2018 by Frost and Sullivan, a global market research firm. Image analysis using deep learning facilities has been the first and the foremost use case for AI.
Oncology is currently the hottest area for clinical application of medical imaging AI. The availability of a large amount of imaging data, coupled with the rising incidence of cancer, is fuelling demand in oncology. Presently, 31 companies are focusing on covering brain tumours, oesophageal, colorectal, liver, lung and prostate cancer, shows the report.
Among the target disease areas, lung cancer, breast cancer, cardiovascular diseases, stroke and neurodegenerative diseases are of prime focus. Paediatrics and orthopaedics are the emerging areas.
In terms of organs, the brain is the most focused for medical imaging AI solutions. Lungs and breast follow.
However, abdominal organs like kidney, liver and prostate have very few solutions focusing on them, says the report.
Applications are predicted to move beyond the current core-imaging modalities and key clinical areas to more challenging, niche and underserved imaging areas in the future.
Big data analytics has gained prominence in the medical imaging arena, critically contributing to the care continuum, along with other electronic health record (EHR) data. The imaging algorithms are capable of deriving metrics using intensive analysis of patterns in a given digital image and detecting specific patterns identified with a specific pathology. Data analytics has been extensively used to complement the analyses made by the radiologist. The future of analytics in diagnostic imaging data may rely on radiology information system (RIS) and picture archiving and communication system (PACS) systems, especially those in the cloud.
Molecular imaging to predict drug response
The advent of molecular medicine has fundamentally changed the treatment of cancer. Molecular imaging has a bigger role to play as personalized medicine seeks
more effective ways to assess tumour response, given the current imaging standards are often inadequate to reliably quantify the changes in the tumour microenvironment. Recent advances in imaging science have made it now possible to visualize previously inaccessible and even unrecognized biological phenomena in cells and tissues in vivo.
The AI space is still mostly focused on whether a machine can recognize a disease condition such as identifying a nodule. However, radiologists are trying to determine what AI can offer and whether AI can make a diagnosis.
Radiomics, which deals with the extraction of data from medical images using algorithms to uncover disease characteristics, is generating a lot of interest as a field of medical study. Studies are underway to find ways to make cancer prognosis and to predict the response to therapy on the basis of imaging.
Molecular brain imaging, using new PET technologies such as non-fluorodeoxyglucose (FDG) PET tracers, is also being explored.
Ultra-high-speed, low dose CTs
Along with developing faster CT scanners, efforts are also ongoing to limit the radiation load to the patients as the CT technology is based on ionizing radiation.
Dual and multi-energy CT enhances image quality by improving material differentiation. Both provide functional information above and beyond CT imaging of morphology alone. Dual-energy CT imaging has several clinical applications.
Siemens Healthineers Somatom Drive CT system is a dual source scanner designed to drive precision in diagnostic imaging with potential to reduce examination time, preparation and follow-up care.
Allowing for more targeted beam focusing, the technology enables CT lung scans at an extremely low dose. It is also useful for spinal diagnostics and orthopaedic examinations. The scanner’s dual energy mode can help clinicians accurately differentiate between tissue and bone. With the system’s ultra-fast scanning speed, the patient’s heart and lung movement do not compromise diagnostic imaging quality, according to a Siemens release.
Multi-energy CT imaging improved clinical differentiations, such as distinguishing blood from calcification and calcification from iodinated contrast. It can also create virtual mono-energetic images for clinical evaluation. Multi-energy CT expands on spectral imaging.
Infervision, a big data and AI company, announced the launch of InferRead CT Chest, a product that detects four different conditions with just one set of chest scans.
InferRead CT Chest will allow a doctor to review an image only once to perform multiple disease screenings in the chest, including lung nodule, chest fractures, bone metastases and bone tumor, chronic lung disease (such as emphysema) and cardiac calcification. The lung nodule screening has been enhanced to provide a complete view of the nodule, including volume and density. This product can automatically compare similar cases from a case report bank to provide further information and diagnostic information to physicians.
In PET-CT, positron emission tomography combines information about the metabolic function with that of the body’s anatomy captured by CT in a single session to provide a more detailed picture of cancerous tissues than either test does alone, with a high level of accuracy.
PET scanning using the tracer fluorine-18 (F-18) fluorodeoxyglucose (FDG), called FDG-PET, is widely used in clinical oncology. PET is also used to diagnose some degenerative brain diseases. Continued development of new radiotracers will lead to a growth in clinical applications for PET/CT both in the field of oncology and in functional neuroimaging such as amyloid PET screening for Alzheimer’s disease etc..
PET-MRI scanners are also being tested on an experimental basis in the clinical setting. PET/MRIs lead to lower radiation exposure compared to a PET-CT. PET-MRIs, however, come with a much higher price tag than PET/CTs. Some studies show that PET/MRI scans of the brain can detect abnormal findings that PET/CT misses in more than 50% of patients scanned.
MRI: Defining pathology
Magnetic resonance imaging (MRI) can detect diseases and pathological tissue. The superior soft tissue contrast in this cutting-edge imaging modality allows better definition of the pathology.
MRI is also increasingly used for guiding, monitoring and controlling percutaneous procedures and surgery. It is billed a as faster and more accurate method of imaging.
More demanding interventional radiology procedures utilize the MRI approach.
Magnetic resonance angiography (MRA), an MRI technique which specifically looks at blood vessels, has been used to image cerebral and renal arteries and other vessels in the head and neck, the aorta and its branches, etc.
Recent MRI allows scans of the lungs. Traditionally, MR imaging has not been possible in the lung since the lungs are filled with air and there is a low density of the hydrogen atoms required to create MR images. Now, Ultrashort Echo Time (UTE) sequence for dedicated pulmonary MRI has been introduced for clinicians to view high susceptibility regions of the lung tissue where signals generally disappear too quickly.
The simultaneous multi-slice application software enables acquisition of MR images simultaneously as opposed to sequentially. With the use of the new software, physicians can bring down the length of MRI brain examinations considerably, which usually vary significantly.
Similarly, cardiac MRI has been made simple with the introduction of new technology. The new MRI software helps automate the image sequences to perform a full 3D chest volume scan, including the full motion of the myocardium during the cardiac cycle and blood flow. It also speeds the imaging time from 70 minutes to about 10 minutes using a single, free-breathing exam, according to reports.
Quantitative diffusion-weighted imaging (DWI) and diffusion tensor imaging (DTI), which offer visualization of the exact location of tumours, is found to be beneficial in the neurosurgical planning and postoperative assessment.
Quantitative imaging is becoming more and more important in clinical practice today, comments Dr C Kesavadas, Professor of Radiology, Sree Chitra Thirunal Institute of Medical Sciences and Technology, Thiruvananthapuram, India. “Differentiation between an infection and a tumour is possible through perfusion technology neuro imaging. Infections and granuloma have mostly low perfusions.”
Clinical researchers have recently demonstrated resting-state functional MRI (fMRI) to develop a prognostic index of clinical response to antipsychotic drug treatment in a cohort of schizophrenia patients. Since clinical response to antipsychotic drug treatment is highly variable, prognostic information can serve as a potential biomarker of treatment response.
The USFDA cleared the first 7 Tesla (7T) MRI device toward the end of 2017. The Magnetom Terra from Siemens more than doubles the static magnetic field strength available. This advanced ultra-high-field scanner is intended for patients over 66 pounds of body weight. The scanner has two coils optimized for clinical neuro and knee imaging. It also features the hyper-fast image reconstruction technology for speeds that are up to 20 times faster than previous generations of 7T research scanners.
New frontiers in real-time US
The use of real-time 3D US imaging has been expanded in scope with improvements in acquisition techniques, reconstruction algorithms, rendering methods and computer GPU acceleration approaches. In obstetrics, abnormalities of the foetal face, rib anomalies and fluid accumulation can be detected. The technology is also used to quantitatively measure left ventricular volume, to diagnose ischemic and congenital heart disease (cardiology), bone erosions in small joints, enthesitis and partial tear of tendons (rheumatology), bladder cancer recurrence and as an alternative to voiding cystourethrography (VCUG) (urology) as well as for surgical guidance and vascular imaging
Foetal HQ heart and vascular software from GE Healthcare for foetal ultrasound now helps evaluate the foetal heart shape, size and contractibility. Radiant Flow shows the blood flow in a 3D view and image slow-flow blood such as neuro-vascular circulation.
Elastography for early detection
Elastography is a newer technique that is based on study findings that a pathological process alters the elastic properties of the involved tissue. When ultrasound is used to assess elastography, it is termed sono-elastography. MRI-elastography uses shear waves to assess tissue displacement in all directions, making it more precise than sonoelastography. The modality is widely used in cases of liver fibrosis, where larger lesions can be easily assessed even in the presence of ascites.
Elastography is very useful in detecting liver cancer. Even the very early changes in the liver tissue, before it starts getting hardened, can be pictured using this imaging modality, said Dr Anbarasu, a consultant radiologist and imaging specialist from Tuticorin, Tamil Nadu.
It is also used to differentiate malignant and benign neoplasms in the breast and in identifying early traumatic changes in muscles and tendons.
X-rays: Yet untapped?
X-ray technology has advanced toward reducing radiation doses and time in the acquisition of scan.
The advent of automatic exposure detection (AED) has transformed radiographic image capture. With the help of AED, digital radiography can be adapted to computed radiography. Wireless DR detector offers advanced image quality, greater reliability, and faster capture speeds.
Tomosynthesis is an advanced application of DR where the X-ray tube sweeps across the patient to get a series of exposures during the pass. The imaging computer draws out a 3D image from the views acquired. Tomosynthesis is presently used in mammography.
Dual-energy imaging is another advance in X-ray technology which is becoming popular. This imaging makes it possible to get three images produced by subtractive software: Just the bones, or just the soft tissues and internal organs, or both together.
Experts say that DR is yet an untapped opportunity and X-ray will find new avenues of clinical usefulness in the future.
“Digital radiology platform has dramatically changed the reporting scenario. Today, images can be sent to the physician sitting anywhere in the world for assessment and further treatment course,” says Dr Anbarasu, consultant radiologist and imaging specialist, Aran Diagnostic Imaging, Coimbatore, Chennai. Teleradiology can be effectively utilized for the people dwelling in the remote areas of India, he added.
Although CTs and MRIs can provide a treasure of information, the planning of the resection line may be difficult when relying on conventional two-dimensional images. 3D imaging techniques are of great value, especially in liver resections. Three-dimensional appearances of liver structures may further improve the results of curative liver surgery. More studies are still needed before 3D becomes a routine clinical procedure.3D models are likely to play an important role in the preoperative planning in many surgical procedures.
New 3D and 4D software increase the contrast of soft tissues and reduce the visibility of metal artifacts compared to traditional CT images.