Artificial intelligence (AI) is rapidly moving from an experimental phase to an implementation phase in many fields, including medicine. The combination of improved availability of large datasets, increasing computing power, and advances in learning algorithms has created major performance breakthroughs in the development of AI applications. In the last 5 years, AI techniques known as deep learning have delivered rapidly improving performance in image recognition, caption generation, and speech recognition. Radiology, in particular, is a prime candidate for early adoption of these techniques. It is anticipated that the implementation of AI in radiology over the next decade will significantly improve the quality, value, and depth of radiology's contribution to patient care and population health, and will revolutionize radiologists' workflows. The Canadian Association of Radiologists (CAR) is the national voice of radiology committed to promoting the highest standards in patient-centered imaging, lifelong learning, and research. The CAR has created an AI working group with the mandate to discuss and deliberate on practice, policy, and patient care issues related to the introduction and implementation of AI in imaging. This white paper provides recommendations for the CAR derived from deliberations between members of the AI working group. This white paper on AI in radiology will inform CAR members and policymakers on key terminology, educational needs of members, research and development, partnerships, potential clinical applications, implementation, structure and governance, role of radiologists, and potential impact of AI on radiology in Canada.

Current deep CNN architectures can be trained with modest-sized medical data sets to achieve clinically useful performance at detecting and excluding common pathology on chest radiographs.

Artificial intelligence (AI) software that analyzes medical images is becoming increasingly prevalent. Unlike earlier generations of AI software, which relied on expert knowledge to identify imaging features, machine learning approaches automatically learn to recognize these features. However, the promise of accurate personalized medicine can only be fulfilled with access to large quantities of medical data from patients. This data could be used for purposes such as predicting disease, diagnosis, treatment optimization, and prognostication. Radiology is positioned to lead development and implementation of AI algorithms and to manage the associated ethical and legal challenges. This white paper from the Canadian Association of Radiologists provides a framework for study of the legal and ethical issues related to AI in medical imaging, related to patient data (privacy, confidentiality, ownership, and sharing); algorithms (levels of autonomy, liability, and jurisprudence); practice (best practices and current legal framework); and finally, opportunities in AI from the perspective of a universal health care system.

No abstract

1. This study examined the inhibitory effects elicited by brain stem stimulation on the somatosensory responses of trigeminal medullary dorsal horn (subnucleus caudalis of the spinal trigeminal nucleus) neurons. Single-unit extracellular recordings were obtained in chloralose-anesthetized cats. Neurons were classified as wide dynamic range (WDR), nociceptive specific (NS), or low-threshold mechanoreceptive (LTM). Conditioning stimuli were delivered to the periaqueductal gray (PAG), nucleus cuneiformis (CU), nucleus raphe magnus (NRM), nucleus reticularis gigantocellularis (NGC), and nucleus reticularis magnocellularis (NMC). 2. Over 97% of the neurons tested could be inhibited by stimulation in all regions except PAG. Stimulation in the PAG inhibited 91% of the neurons tested. There was no statistically significant difference in the incidence of inhibition of WDR and NS nociceptive (noci) neurons and the LTM nonnociceptive (nonnoci) neurons. 3. Mean stimulation intensities necessary to produce inhibition were determined for each neuron from each stimulation site. The current thresholds necessary to inhibit the responses of noci neurons were found to be significantly lower, on the average, than those of nonnoci neurons at stimulation sites in the PAG, CU, and NGC. 4. Inhibition of the responses of WDR neurons required a lower mean current than for NS neurons but was statistically significant only for PAG and NGC. Thresholds for inhibiting the responses of NS neurons were similar to those for inhibiting the responses of LTM neurons for all regions except CU, where LTM thresholds were markedly but not significantly higher. 5. Stimulation thresholds were found to be lowest in NMC, while in NGC, NRM, and CU they were all similar and slightly higher. Stimulation in the PAG required the highest currents to produce inhibition. 6. These results indicate that stimulation in NRM and PAG not only inhibits the responses of noci neurons but also those of nonnoci neurons. Furthermore, stimulation in reticular regions adjacent to NRM and PAG is frequently even more effective in inhibiting the responses of both noci and nonnoci neurons. In addition, WDR neurons are more effectively inhibited than NS or LTM neurons. These results are compared with those obtained using similar methods in cat lumbar dorsal horn.

Automated methods of radiation dose data collection permit a detailed analysis of radiation dose according to protocol and equipment over time. Radiation dose optimization measures were effective, but their full value may be realized only with changes in internal processes and real-time, prospective data monitoring and analysis.

No abstract

We examined the patient and physician characteristics related to the use and yield of computed tomography pulmonary angiogram (CTPA) for the diagnosis of pulmonary embolism (PE) at a tertiary academic hospital emergency department (ED). A cross-sectional retrospective study was conducted on 835 consecutive ED patients with suspected PE who underwent CTPA. Radiology report data were extracted from our institution's RIS PACS software (Syngo Imaging, Siemens) based on a targeted search of all CTPA reports from 2010 to 2012. Utilization and PE positivity rates of CTPA were calculated and correlated with patient characteristics including age and gender, as well as emergency physician (EP) characteristics including gender, years in practice, and training certification. Acute PE was diagnosed in 17.8 % of patients. A further 32.9 % of the scans were negative for PE but had other clinically significant findings. We found higher utilization rates in female and older patients (p < 0.001), however, without corresponding differences in PE positivity rates compared to their male and younger counterparts. There was a high inter-physician variation in CTPA utilization rate (range 0.21-0.77 scans per 100 patients seen) and PE positivity rate (range 6.7-38.9 %). However, neither rates correlated with EP years of experience (p > 0.15 with cut-offs at 5, 10, and 20 years post-residency), gender (p = 0.59), or training certification (p = 0.56 between EPs certified by the 5-year program of the Royal College of Physicians of Canada versus the 3-year program of the College of Family Physicians of Canada). Our study demonstrated considerable inter-physician variability in the utilization and PE positivity rates of CTPA. These results suggest an opportunity for a more standardized approach to the use of CTPA among EPs at our institution.