Unveiling COVID-19 from CHEST X-Ray with Deep Learning: A Hurdles Race with Small Data.

The possibility to use widespread and simple chest X-ray (CXR) imaging for early screening of COVID-19 patients is attracting much interest from both the clinical and the AI community. In this study we provide insights and also raise warnings on what is reasonable to expect by applying deep learning to COVID classification of CXR images. We provide a methodological guide and critical reading of an extensive set of statistical results that can be obtained using currently available datasets. In particular, we take the challenge posed by current small size COVID data and show how significant can be the bias introduced by transfer-learning using larger public non-COVID CXR datasets. We also contribute by providing results on a medium size COVID CXR dataset, just collected by one of the major emergency hospitals in Northern Italy during the peak of the COVID pandemic. These novel data allow us to contribute to validate the generalization capacity of preliminary results circulating in the scientific community. Our conclusions shed some light into the possibility to effectively discriminate COVID using CXR.

Prostate cancer detection with biparametric magnetic resonance imaging (bpMRI) by readers with different experience: performance and comparison with multiparametric (mpMRI).

PURPOSE: To study the detection of clinically significant prostate cancer (PCa) by readers with different experience, comparing performance with biparametric magnetic resonance imaging (bmMRI) and with the reference multiparametric (mpMRI). METHODS: Retrospective analysis of 68 patients with mpMRI of the prostate at 1.5 Tesla using a 32 phased-array coil. Forty-five patients (cases) underwent radical prostatectomy, whereas 23 (controls) had a negative prostate biopsy and ≥ 2.5 years of negative follow-up. Six observers (two with 1000 cases interpreted, two with 300, two with 100) performed the analysis first with bpMRI including diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC) maps and T2-weighted (T2W) imaging in three planes and, after 1 month, with mpMRI, adding dynamic contrast enhancement (DCE). The performance was quantified by sensitivity (SNS), specificity (SPC) and area under the curve (AUC) of the ROC (Receiver Operating Characteristics) procedure. RESULTS: Concordance within observers of equivalent experience was good (weighted Cohen's k ≈ 0.7). The two expert readers performed as well in bpMRI as in mpMRI (SNS = 0.91-0.96, AUC = 0.86-0.93; p ≥ 0.10); readers with 300 cases performed well in mpMRI, but significantly worse in bpMR: SNS = 0.58 versus 0.91 (p < 0.0001) and AUC = 0.73 versus 0.86 (p = 0.01); the limited experience of readers with 100 cases showed in mpMRI (SNS = 0.71; AUC = 0.77) and even more in bpMRI (SNS = 0.50; AUC = 0.68). CONCLUSION: The study revealed the impact of the readers' experience when using bpMRI. The bpMRI without contrast media was a valid alternative for expert readers, whereas less experienced ones needed DCE to significantly boost SNS and AUC. Results indicate 700-800 cases as threshold for reliable interpretation with bpMRI.

Association Between Haptoglobin Phenotype and Microvascular Obstruction in Patients With STEMI: A Cardiac Magnetic Resonance Study.

OBJECTIVES: This study aimed to evaluate the correlation between different haptoglobin (Hp) phenotypes and myocardial infarction characteristics as detected by cardiac magnetic resonance (CMR) in consecutive patients after ST-segment elevation myocardial infarction (STEMI). BACKGROUND: Hp is a plasma protein that prevents iron-mediated oxidative tissue damage. CMR has emerged as the gold standard technique to detect left ventricular ejection fraction (LVEF), extent of scar with late gadolinium enhancement (LGE) technique, microvascular obstruction (MVO), and myocardial hemorrhage (MH) in patients with STEMI treated by primary percutaneous coronary intervention (pPCI). METHODS: A total of 145 consecutive STEMI patients (mean age 62.2 ± 10.3 years; 78% men) were prospectively enrolled and underwent Hp phenotyping and CMR assessment within 1 week after STEMI. RESULTS: CMR showed an area at risk (AAR) involving 26.6 ± 19.1% of left ventricular (LV) mass with a late LGE extent of 15.2 ± 13.1% of LV mass. MVO and MH occurred in 38 (26%) and 12 (8%) patients, respectively. Hp phenotypes 1-1, 2-1, 2-2 were observed in 15 (10%), 62 (43%), and 68 (47%), respectively. Multivariable analysis demonstrated that body mass index, Hp2-2, diabetes, and peak troponin I were independent predictors of MVO with Hp2-2 associated with the highest odds ratio (OR) (OR: 5.5 [95% confidence interval [CI]: 2.1 to 14.3; p < 0.001]). Hp2-2 significantly predicted both the presence (area under the curve [AUC]: 0.63 [95% CI: 0.53 to 0.72; p = 0.008]) and extent of MVO (AUC: 0.63 [95% CI: 0.54 to 0.72; p = 0.007]). CONCLUSIONS: Hp phenotype is an independent predictor of MVO. Therefore, Hp phenotyping could be used for risk stratification and may be useful in assessing new therapies to reduce myocardial reperfusion injury in patients with STEMI.

Incremental Diagnostic Value of Stress Computed Tomography Myocardial Perfusion With Whole-Heart Coverage CT Scanner in Intermediate- to High-Risk Symptomatic Patients Suspected of Coronary Artery Disease.

OBJECTIVES: The goal of this study was to evaluate the diagnostic accuracy of stress computed tomography myocardial perfusion (CTP) for the detection of functionally significant coronary artery disease (CAD) by using invasive coronary angiography (ICA) plus invasive fractional flow reserve (FFR) as the reference standard in consecutive intermediate- to high-risk symptomatic patients. BACKGROUND: Stress CTP recently emerged as a potential strategy to combine the anatomic and functional evaluation of CAD in a single scan. METHODS: A total of 100 consecutive symptomatic patients scheduled for ICA were prospectively enrolled. All patients underwent rest coronary computed tomography angiography (CTA) followed by stress static CTP with a whole-heart coverage CT scanner (Revolution CT, GE Healthcare, Milwaukee, Wisconsin). Diagnostic accuracy and overall effective dose were assessed and compared versus those of ICA and invasive FFR. RESULTS: The prevalence of obstructive CAD and functionally significant CAD were 69% and 44%, respectively. Coronary CTA alone demonstrated a per-vessel and per-patient sensitivity, specificity, negative predictive value, positive predictive value, and accuracy of 98%, 76%, 99%, 63%, and 83% and of 98%, 54%, 96%, 68%, and 76%, respectively. Combining coronary CTA with stress CTP, per-vessel and per-patient sensitivity, specificity, negative predictive value, positive predictive value, and accuracy were 91%, 94%, 96%, 86%, and 93% and 98%, 83%, 98%, 86%, and 91%, with a significant improvement in specificity, positive predictive value, and accuracy in both models. The mean effective dose for coronary CTA and stress CTP were 2.8 ± 1.4 mSv and 2.5 ± 1.1 mSv. CONCLUSIONS: The inclusion of stress CTP for the evaluation of patients with an intermediate to high risk for CAD is feasible and improved the diagnostic performance of coronary CTA for detecting functionally significant CAD.

Quantitative vs. qualitative evaluation of static stress computed tomography perfusion to detect haemodynamically significant coronary artery disease.

Aims: To compare the diagnostic accuracy of stress computed tomography myocardial perfusion (CTP) for the detection of significant coronary artery disease with visual approach vs. quantitative analysis with transmural perfusion ratio (TPR) in consecutive symptomatic patients scheduled for invasive coronary angiography (ICA) plus invasive fractional flow reserve (FFR). Methods and results: Eighty-eight consecutive symptomatic patients underwent rest coronary computed tomography angiography (cCTA) followed by static stress-CTP. Diagnostic accuracy of cCTA + stress-CTP with visual evaluation and with TPR measurement was calculated and compared with ICA and invasive FFR. Addition of stress-CTP with qualitative evaluation to rest-cCTA showed sensitivity, specificity, negative and positive predictive values, and accuracy at a vessel and patient level of 92%, 92%, 97%, 82%, 92% and 98%, 80%, 97%, 82%, 89%, respectively indicating a significant improvement of specificity, positive predictive value, and accuracy values vs. rest-cCTA in both models. Similarly, addition of stress-CTP with TPR evaluation to rest-cCTA showed sensitivity, specificity, negative and positive predictive values, and accuracy at a vessel and patient level of 84%, 90%, 93%, 76%, 88% and 91%, 71%, 89%, 75%, 81%, respectively indicating a significant improvement of specificity, positive predictive value values vs. rest-cCTA only in a vessel-based model and of positive predictive value in a patient-based model. When cCTA + stress-CTP with qualitative evaluation was compared with cCTA + stress-CTP with TPR estimation, no differences were found in terms of diagnostic performance. Conclusion: The addition of stress-CTP with visual evaluation to cCTA imaging has similar diagnostic performance when compared with the quantitative analysis of myocardial perfusion based on TPR measurement.