Deep learning workflow in radiology: a primer Montagnon, Emmanuel; Cerny, Milena; Cadrin-Chênevert, Alexandre ...
Insights into imaging,
02/2020, Letnik:
11, Številka:
1
Journal Article
Recenzirano
Odprti dostop
Interest for deep learning in radiology has increased tremendously in the past decade due to the high achievable performance for various computer vision tasks such as detection, segmentation, ...classification, monitoring, and prediction. This article provides step-by-step practical guidance for conducting a project that involves deep learning in radiology, from defining specifications, to deployment and scaling. Specifically, the objectives of this article are to provide an overview of clinical use cases of deep learning, describe the composition of multi-disciplinary team, and summarize current approaches to patient, data, model, and hardware selection. Key ideas will be illustrated by examples from a prototypical project on imaging of colorectal liver metastasis. This article illustrates the workflow for liver lesion detection, segmentation, classification, monitoring, and prediction of tumor recurrence and patient survival. Challenges are discussed, including ethical considerations, cohorting, data collection, anonymization, and availability of expert annotations. The practical guidance may be adapted to any project that requires automated medical image analysis.
Objectives
To perform head-to-head comparisons of the feasibility and diagnostic performance of transient elastography (TE), point shear-wave elastography (pSWE), and magnetic resonance elastography ...(MRE).
Methods
This prospective, cross-sectional, dual-center imaging study included 100 patients with known or suspected chronic liver disease caused by hepatitis B or C virus, nonalcoholic fatty liver disease, or autoimmune hepatitis identified between 2014 and 2018. Liver stiffness measured with the three elastographic techniques was obtained within 6 weeks of a liver biopsy. Confounding effects of inflammation and steatosis on association between fibrosis and liver stiffness were assessed. Obuchowski scores and AUCs for staging fibrosis were evaluated and the latter were compared using the DeLong method.
Results
TE, pSWE, and MRE were technically feasible and reliable in 92%, 79%, and 91% subjects, respectively. At univariate analysis, liver stiffness measured by all techniques increased with fibrosis stages and inflammation and decreased with steatosis. For classification of dichotomized fibrosis stages, the AUCs were significantly higher for distinguishing stages F0 vs. ≥ F1 with MRE than with TE (0.88 vs. 0.71;
p
< 0.05) or pSWE (0.88 vs. 0.73;
p
< 0.05), and for distinguishing stages ≤ F1 vs. ≥ F2 with MRE than with TE (0.85 vs. 0.75;
p
< 0.05). TE, pSWE, and MRE Obuchowski scores for staging fibrosis stages were respectively 0.89 (95% CI 0.85–0.93), 0.90 (95% CI 0.85–0.94), and 0.94 (95% CI 0.91–0.96).
Conclusion
MRE provided a higher diagnostic performance than TE and pSWE for staging early stages of liver fibrosis.
Trial registration
NCT02044523
Key Points
• The technical failure rate was similar between MRE and US-based elastography techniques.
• Liver stiffness measured by MRE and US-based elastography techniques increased with fibrosis stages and inflammation and decreased with steatosis.
• MRE provided a diagnostic accuracy higher than US-based elastography techniques for staging of early stages of histology-determined liver fibrosis.
Integrase strand transfer inhibitors (INSTIs) are recommended for first-line HIV therapy based on their relatively high genetic barrier to resistance. Although raltegravir (RAL) and elvitegravir ...(EVG) resistance profiles are well-characterized, resistance patterns for dolutegravir (DTG), bictegravir (BIC), and cabotegravir (CAB) remain largely unknown. Here, in vitro drug selections compared the development of resistance to DTG, BIC, CAB, EVG and RAL using clinical isolates from treatment-naïve primary HIV infection (PHI) cohort participants (n = 12), and pNL4.3 recombinant strains encoding patient-derived Integrase with (n = 5) and without (n = 5) the E157Q substitution.
Patient-derived viral isolates were serially passaged in PHA-stimulated cord blood mononuclear cells in the presence of escalating concentrations of INSTIs over the course of 36-46 weeks. Drug resistance arose more rapidly in primary clinical isolates with EVG (12/12), followed by CAB (8/12), DTG (8/12) and BIC (6/12). For pNL4.3 recombinant strains encoding patient-derived integrase, the comparative genetic barrier to resistance was RAL > EVG > CAB > DTG and BIC. The E157Q substitution in integrase delayed the advent of resistance to INSTIs. With EVG, T66I/A, E92G/V/Q, T97A or R263K (n = 16, 3, 2 and 1, respectively) arose by weeks 8-16, followed by 1-4 accessory mutations, conferring high-level resistance (> 100-fold) by week 36. With DTG and BIC, solitary R263K (n = 27), S153F/Y (n = 7) H51Y (n = 2), Q146 R (n = 3) or S147G (n = 1) mutations conferred low-level (< 3-fold) resistance at weeks 36-46. Similarly, most CAB selections (n = 18) resulted in R263K, S153Y, S147G, H51Y, or Q146L solitary mutations. However, three CAB selections resulted in Q148R/K followed by secondary mutations conferring high-level cross-resistance to all INSTIs. EVG-resistant viruses (T66I/R263K, T66I/E157Q/R263K, and S153A/R263K) retained residual susceptibility when switched to DTG, BIC or CAB, losing T66I by week 27. Two EVG-resistant variants developed resistance to DTG, BIC and CAB through the additional acquisition of E138A/Q148R and S230N, respectively. One EVG-resistant variant (T66I) acquired L74M/G140S/S147G, L74M/E138K/S147G and H51Y with DTG CAB and BIC, respectively.
Second generation INSTIs show a higher genetic barrier to resistance than EVG and RAL. The potency of CAB was lower than BIC and DTG. The development of Q148R/K with CAB can result in high-level cross-resistance to all INSTIs.
The management of localized rectal cancer has evolved significantly over the last two years. On one hand, intensification of treatments (radio-chemotherapy, chemotherapy, then surgery) for the most ...advanced tumors has shown an improvement in clinical results compared to less intense regiments. On the other hand, the possibility, as for prostate cancers, of opting for active surveillance without surgery in patients presenting a complete clinical response after a treatment phase, is now accepted. More recently, the Swiss recommendations for the surveillance of rectal cancer have been modified and now differ from those of colon cancers, by incorporating pelvic MRI and rectoscopy in addition, as well as special guidelines for tumors under active surveillance.
Viral phylogenetics revealed two patterns of HIV-1 spread among MSM in Quebec. While most HIV-1 strains ( n = 2011) were associated with singleton/small clusters (cluster size 1-4), 30 viral ...lineages formed large networks (cluster size 20-140), contributing to 42% of diagnoses between 2011 and 2015. Herein, tissue culture selections ascertained if large cluster lineages possessed higher replicative fitness than singleton/small cluster isolates, allowing for viral escape from integrase inhibitors.
Primary HIV-1 isolates from large 20+ cluster ( n = 11) or singleton/small cluster ( n = 6) networks were passaged in vitro in escalating concentrations of dolutegravir, elvitegravir and lamivudine for 24-36 weeks. Sanger and deep sequencing assessed genotypic changes under selective drug pressure.
Large cluster HIV-1 isolates selected for resistance to dolutegravir, elvitegravir and lamivudine faster than HIV-1 strains forming small clusters. With dolutegravir, large cluster HIV-1 variants acquired solitary R263K ( n = 7), S153Y ( n = 1) or H51Y ( n = 1) mutations as the dominant quasi-species within 8-12 weeks as compared with small cluster lineages where R263K ( n = 1/6), S153Y (1/6) or WT species (4/6) were observed after 24 weeks. Interestingly, dolutegravir-associated mutations compromised viral replicative fitness, precluding escalations in concentrations beyond 5-10 nM. With elvitegravir, large cluster variants more rapidly acquired first mutations (T66I, A92G, N155H or S147G) by week 8 followed by sequential accumulation of multiple mutations leading to viral escape (>10 μM) by week 24.
Further studies are needed to understand virological features of large cluster viruses that may favour their transmissibility, replicative competence and potential to escape selective antiretroviral drug pressure.
To determine whether serum magnesium and calcium concentrations are causally associated with ischemic stroke or any of its subtypes using the mendelian randomization approach.
Analyses were conducted ...using summary statistics data for 13 single-nucleotide polymorphisms robustly associated with serum magnesium (n = 6) or serum calcium (n = 7) concentrations. The corresponding data for ischemic stroke were obtained from the MEGASTROKE consortium (34,217 cases and 404,630 noncases).
In standard mendelian randomization analysis, the odds ratios for each 0.1 mmol/L (about 1 SD) increase in genetically predicted serum magnesium concentrations were 0.78 (95% confidence interval CI 0.69-0.89;
= 1.3 × 10
) for all ischemic stroke, 0.63 (95% CI 0.50-0.80;
= 1.6 × 10
) for cardioembolic stroke, and 0.60 (95% CI 0.44-0.82;
= 0.001) for large artery stroke; there was no association with small vessel stroke (odds ratio 0.90, 95% CI 0.67-1.20;
= 0.46). Only the association with cardioembolic stroke was robust in sensitivity analyses. There was no association of genetically predicted serum calcium concentrations with all ischemic stroke (per 0.5 mg/dL about 1 SD increase in serum calcium: odds ratio 1.03, 95% CI 0.88-1.21) or with any subtype.
This study found that genetically higher serum magnesium concentrations are associated with a reduced risk of cardioembolic stroke but found no significant association of genetically higher serum calcium concentrations with any ischemic stroke subtype.