RESUMO
The application of big data, radiomics, machine learning, and artificial intelligence (AI) algorithms in radiology requires access to large data sets containing personal health information. Because machine learning projects often require collaboration between different sites or data transfer to a third party, precautions are required to safeguard patient privacy. Safety measures are required to prevent inadvertent access to and transfer of identifiable information. 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 Ethical and Legal standing committee with the mandate to guide the medical imaging community in terms of best practices in data management, access to health care data, de-identification, and accountability practices. Part 2 of this article will inform CAR members on the practical aspects of medical imaging de-identification, strengths and limitations of de-identification approaches, list of de-identification software and tools available, and perspectives on future directions.
Assuntos
Inteligência Artificial/ética , Anonimização de Dados/ética , Diagnóstico por Imagem/ética , Radiologistas/ética , Algoritmos , Canadá , Humanos , Aprendizado de Máquina , Sociedades MédicasRESUMO
The application of big data, radiomics, machine learning, and artificial intelligence (AI) algorithms in radiology requires access to large data sets containing personal health information. Because machine learning projects often require collaboration between different sites or data transfer to a third party, precautions are required to safeguard patient privacy. Safety measures are required to prevent inadvertent access to and transfer of identifiable information. 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 Ethical and Legal standing committee with the mandate to guide the medical imaging community in terms of best practices in data management, access to health care data, de-identification, and accountability practices. Part 1 of this article will inform CAR members on principles of de-identification, pseudonymization, encryption, direct and indirect identifiers, k-anonymization, risks of reidentification, implementations, data set release models, and validation of AI algorithms, with a view to developing appropriate standards to safeguard patient information effectively.
Assuntos
Inteligência Artificial/ética , Anonimização de Dados/ética , Diagnóstico por Imagem/ética , Radiologistas/ética , Algoritmos , Canadá , Humanos , Aprendizado de Máquina , Sociedades MédicasRESUMO
A 36-year-old woman presented at 16 weeks' gestation with severe hypertension. In comparison to the non-pregnant reference normal ranges, potassium was 3.1-3.9 mmol/L, aldosterone 2570-3000 pmol/L (N 250-2885) renin was unsuppressed (24-76.4 ng/L (N1.7-23.9)), with aldosterone to renin ratios in the reference range. An adrenal MRI scan demonstrated a 1.8 × 1.4 cm left adrenal adenoma. Primary aldosteronism was strongly suspected and surgery considered. However, she was managed conservatively with labetalol and modified-release nifedipine with no obstetric complications. Post-partum blood pressures remained elevated with normal aldosterone (539 pmol/L), unsuppressed renin (5.2 ng/L) and normal aldosterone-to-renin ratio (104 (N < 144)). Suspected primary hyperaldosteronism is challenging to investigate and manage in pregnancy. The accepted screening and confirmatory tests are either contraindicated or not validated in pregnancy. Pregnancy has significant effects on the renin-angiotensin-aldosterone pathway leading to physiologic elevations in both aldosterone and renin. While primary hyperaldosteronism has been associated with poor pregnancy outcomes, optimal management in pregnancy is not clearly established.
RESUMO
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.
Assuntos
Inteligência Artificial/ética , Inteligência Artificial/legislação & jurisprudência , Radiologia/ética , Radiologia/legislação & jurisprudência , Canadá , Humanos , Guias de Prática Clínica como Assunto , Radiologistas/ética , Radiologistas/legislação & jurisprudência , Sociedades MédicasAssuntos
Meios de Contraste/efeitos adversos , Gadolínio/efeitos adversos , Globo Pálido/efeitos dos fármacos , Imageamento por Ressonância Magnética , Transtornos Parkinsonianos/induzido quimicamente , Risco , Idoso , Feminino , Humanos , Imageamento por Ressonância Magnética/estatística & dados numéricos , MasculinoRESUMO
Automatic vertebra recognition, including the identification of vertebra locations and naming in multiple image modalities, are highly demanded in spinal clinical diagnoses where large amount of imaging data from various of modalities are frequently and interchangeably used. However, the recognition is challenging due to the variations of MR/CT appearances or shape/pose of the vertebrae. In this paper, we propose a method for multi-modal vertebra recognition using a novel deep learning architecture called Transformed Deep Convolution Network (TDCN). This new architecture can unsupervisely fuse image features from different modalities and automatically rectify the pose of vertebra. The fusion of MR and CT image features improves the discriminativity of feature representation and enhances the invariance of the vertebra pattern, which allows us to automatically process images from different contrast, resolution, protocols, even with different sizes and orientations. The feature fusion and pose rectification are naturally incorporated in a multi-layer deep learning network. Experiment results show that our method outperforms existing detection methods and provides a fully automatic location+naming+pose recognition for routine clinical practice.
Assuntos
Aprendizado de Máquina , Coluna Vertebral/diagnóstico por imagem , Automação , HumanosRESUMO
BACKGROUND: International Classification of Diseases, 10th Revision codes (ICD-10) for autosomal dominant polycystic kidney disease (ADPKD) is used within several administrative health care databases. It is unknown whether these codes identify patients who meet strict clinical criteria for ADPKD. OBJECTIVE: The objective of this study is (1) to determine whether different ICD-10 coding algorithms identify adult patients who meet strict clinical criteria for ADPKD as assessed through medical chart review and (2) to assess the number of patients identified with different ADPKD coding algorithms in Ontario. DESIGN: Validation study of health care database codes, and prevalence. SETTING: Ontario, Canada. PATIENTS: For the chart review, 201 adult patients with hospital encounters between April 1, 2002, and March 31, 2014, assigned either ICD-10 codes Q61.2 or Q61.3. MEASUREMENTS: This study measured positive predictive value of the ICD-10 coding algorithms and the number of Ontarians identified with different coding algorithms. METHODS: We manually reviewed a random sample of medical charts in London, Ontario, Canada, and determined whether or not ADPKD was present according to strict clinical criteria. RESULTS: The presence of either ICD-10 code Q61.2 or Q61.3 in a hospital encounter had a positive predictive value of 85% (95% confidence interval [CI], 79%-89%) and identified 2981 Ontarians (0.02% of the Ontario adult population). The presence of ICD-10 code Q61.2 in a hospital encounter had a positive predictive value of 97% (95% CI, 86%-100%) and identified 394 adults in Ontario (0.003% of the Ontario adult population). LIMITATIONS: (1) We could not calculate other measures of validity; (2) the coding algorithms do not identify patients without hospital encounters; and (3) coding practices may differ between hospitals. CONCLUSIONS: Most patients with ICD-10 code Q61.2 or Q61.3 assigned during their hospital encounters have ADPKD according to the clinical criteria. These codes can be used to assemble cohorts of adult patients with ADPKD and hospital encounters.
MISE EN CONTEXTE: La 10e révision des codes de l'International Classification of Diseases (ICD-10) est utilisée dans plusieurs bases de données administratives des centres de soins pour le classement de la maladie polykystique autosomique dominante (MPR). On ignore toutefois si ces codes permettent d'identifier clairement les patients qui satisfont les critères cliniques stricts de la maladie. OBJECTIFS DE L'ÉTUDE: 1) Déterminer si les différents algorithmes de codage de la ICD-10 réussissent à identifier de manière efficace les patients adultes satisfaisant les critères cliniques stricts de la MPR tels qu'évalués par la consultation des dossiers médicaux; 2) Évaluer le nombre de patients qui sont identifiés par les différents algorithmes de codage pour la MPR, en Ontario. CADRE ET TYPE D'ÉTUDE: Il s'agit d'une étude de validation des codes de classification obtenus dans les bases de données des centres de soins de l'Ontario, au Canada, ainsi que de leur prévalence. PATIENTS: On a révisé les dossiers médicaux de 201 patients adultes ayant reçu une consultation en centre hospitalier entre le 1er avril 2002 et le 31 mars 2014, et à qui les codes ICD-10 Q61.2 ou Q61.3 pour la MPR ont été assignés. MESURES: Les valeurs prédictives positives des algorithmes de codage ICD-10 ainsi que le nombre d'Ontariens identifiés comme patients atteints de MPR par les différents algorithmes de codage ont été retenus pour l'étude. MÉTHODOLOGIE: Un échantillon aléatoire de dossiers médicaux en provenance de London, en Ontario (Canada) a été révisé manuellement afin de déterminer lesquels indiquaient la présence d'une MPR selon les critères cliniques stricts pour cette maladie. RÉSULTATS: La présence des codes ICD-10 Q61.2 ou Q61.3 lors d'une consultation à l'hôpital a eu une valeur prédictive positive dans 85% des cas (IC 95%: 79 à 89%), et a permis l'identification d'un total de 2 981 patients ontariens (0,02% de la population adulte en Ontario). Le codage ICD-10 Q61.2 à lui seul a eu une valeur prédictive positive dans 97% des cas (IC 95%: 86 à 100%) et a permis l'identification de 394 patients (0,003% de la population adulte en Ontario). LIMITES DE L'ÉTUDE: 1) Nous n'avons pu calculer aucune autre mesure de validité; 2) Les algorithmes de codage n'identifient pas les patients s'ils ne sont pas en consultation en centre hospitalier; 3) Les pratiques de codage peuvent varier d'un hôpital à un autre. CONCLUSIONS: La majorité des patients codés ICD-10 Q61.2 ou Q61.3 à la suite d'une consultation en centre hospitalier était atteinte de maladie polykystique autosomique dominante selon les critères cliniques stricts pour cette maladie. Ainsi, cette codification peut être utilisée pour jumeler des cohortes de patients adultes atteints de MPR avec leurs consultations en hôpital.
RESUMO
OBJECTIVES: Transcutaneous bowel sonography is a nonionizing imaging modality used in inflammatory bowel disease. Although available in Europe, its uptake in North America has been limited. Since the accuracy of bowel sonography is highly operator dependent, low-volume centers in North America may not achieve the same diagnostic accuracy reported in the European literature. Our objective was to determine the diagnostic accuracy of bowel sonography in a nonexpert low-volume center. METHODS: All cases of bowel sonography at a single tertiary care center during an 18-month period were reviewed. Bowel sonography was compared with reference standards, including small-bowel follow-through, computed tomography, magnetic resonance imaging, colonoscopy, and surgical findings. RESULTS: A total of 103 cases were included for analysis during the study period. The final diagnoses included Crohn disease (72), ulcerative colitis (8), hemolytic uremic syndrome (1), and normal (22). The sensitivity and specificity of bowel sonography for intestinal wall inflammation were 87.8% and 92.6%, respectively. In the subset of patients who had complications of Crohn disease, the sensitivity and specificity were 50% and 100% for fistulas and 14% and 100% for strictures. One patient had an abscess, which was detected by bowel sonography. Abnormal bowel sonographic findings contributed to the escalation of treatment in 55% of cases. CONCLUSIONS: Bowel sonography for inflammatory bowel disease can be performed in low-volume centers and provides diagnostic accuracy for luminal disease comparable with published data, although it is less sensitive for complications of Crohn disease.