Consistency of convolutional neural networks in dermoscopic melanoma recognition: A prospective real-world study about the pitfalls of augmented intelligence

BACKGROUND

Deep-learning convolutional neural networks (CNNs) have outperformed even experienced dermatologists in dermoscopic melanoma detection under controlled conditions. It remains unexplored how real-world dermoscopic image transformations affect CNN robustness.

OBJECTIVES

To investigate the consistency of melanoma risk assessment by two commercially available CNNs to help formulate recommendations for current clinical use.

METHODS

A comparative cohort study was conducted from January to July 2022 at the Department of Dermatology, University Hospital Basel. Five dermoscopic images of 116 different lesions on the torso of 66 patients were captured consecutively by the same operator without deliberate rotation. Classification was performed by two CNNs (CNN-1/CNN-2). Lesions were divided into four subgroups based on their initial risk scoring and clinical dignity assessment. Reliability was assessed by variation and intraclass correlation coefficients. Excisions were performed for melanoma suspicion or two consecutively elevated CNN risk scores, and benign lesions were confirmed by expert consensus (n = 3).

RESULTS

117 repeated image series of 116 melanocytic lesions (2 melanomas, 16 dysplastic naevi, 29 naevi, 1 solar lentigo, 1 suspicious and 67 benign) were classified. CNN-1 demonstrated superior measurement repeatability for clinically benign lesions with an initial malignant risk score (mean variation coefficient (mvc): CNN-1: 49.5(±34.3)%; CNN-2: 71.4(±22.5)%; p = 0.03), while CNN-2 outperformed for clinically benign lesions with benign scoring (mvc: CNN-1: 49.7(±22.7)%; CNN-2: 23.8(±29.3)%; p = 0.002). Both systems exhibited lowest score consistency for lesions with an initial malignant risk score and benign assessment. In this context, averaging three initial risk scores achieved highest sensitivity of dignity assessment (CNN-1: 94%; CNN-2: 89%). Intraclass correlation coefficients indicated 'moderate'-to-'good' reliability for both systems (CNN-1: 0.80, 95% CI:0.71-0.87, p < 0.001; CNN-2: 0.67, 95% CI:0.55-0.77, p < 0.001).

CONCLUSIONS

Potential user-induced image changes can significantly influence CNN classification. For clinical application, we recommend using the average of three initial risk scores. Furthermore, we advocate for CNN robustness optimization by cross-validation with repeated image sets.

TRIAL REGISTRATION

ClinicalTrials.gov (NCT04605822).

Hypothermic preservation of lung allograft inhibits cytokine-induced chemoattractant-1, endothelial leucocyte adhesion molecule, vascular cell adhesion molecule-1 and intracellular adhesion molecule-1 expression

Organ dysfunction is a major clinical problem after lung transplantation. Prolonged cold ischaemia and reperfusion injury are believed to play a central role in this complication. The influence of cold preservation on subsequent warm reperfusion was studied in an isolated, ventilated and perfused rat lung. Rat lungs were flushed with cold Perfadex-solution and stored at 4 degrees C for different time periods. Thereafter lungs were perfused and ventilated for up to 3 h. Physiological parameters, production of inflammatory mediators and leucocyte infiltration were measured before and after perfusion. Lungs subjected to a cold ischaemia time of up to 6 h showed stable physiological conditions when perfused for 3 h. However, cold-ischaemia time beyond 6 h resulted in profound tissue oedema, thereby impairing ventilation and perfusion. Warm reperfusion and ventilation per se induced a strong inflammatory response, as demonstrated by a significant up-regulation of chemokines and adhesion molecules (cytokine-induced chemoattractant-1, intracellular adhesion molecule and endothelial leucocyte adhesion molecule), accompanied by enhanced leucocyte infiltration. Although the up-regulation of inflammatory mediators was blunted in lungs that were subjected to cold ischaemia, this did not influence leucocyte infiltration. In fact, cold ischaemia time correlated with leucocyte sequestration. Although cold preservation inhibits the expression of inflammatory mediators it does not affect leucocyte sequestration during warm reperfusion. Cold preservation might cause impairment of the endothelial barrier function, as evidenced by tissue oedema and profound leucocyte infiltration.

Level of supplemental protein does not influence the ruminally undegradable protein value

Two experiments were conducted to determine whether elevating the percentage of ruminally undegradable protein (RUP) in the diet would influence the RUP value of the protein feedstuff. A single-effluent, continuous-culture study was designed to test the effect of RUP inclusion rate in the diet on ruminal degradability of the protein. Treatments consisted (DM basis) of a control diet with no supplemental protein, control + 2.5% bloodmeal (BM-L), control + 5% bloodmeal (BM-H), control + 4.45% soybean meal (SBM-L), and control + 8.89% soybean meal (SBM-H). Proteolytic activity and total VFA concentration were not affected (P = 0.73 and P = 0.13) by treatment. Within protein source, dietary RUP value was not affected (P = 0.94) by level of inclusion. When corrected for control diet RUP flow, the RUP value of the blood meal (BM) protein was higher (P = 0.01) than soybean meal (SBM); however, level of supplementation did not affect (P = 0.07) the RUP value of BM or SBM. In Exp. 2, 32 British x Continental crossbred steers (276 +/- 26.3 kg) were fed for 72 d to examine the effects of balancing the AA:energy ratio, using BM as a RUP source, on ADG, G:F, and lean tissue deposition. Diets were formulated to provide increasing levels of arginine, while ruminally degradable protein and energy were held constant. Four dietary treatments provided 0.5, 1, 1.5, and 2x the required amount of arginine, whereas the control diet had no BM included. Daily DMI averaged 7.6 kg/steer and did not differ (P = 0.71) among treatments. Steers gained an average of 1.9 kg/d and average G:F was 0.260, with no differences (P = 0.60 and P = 0.97, respectively) among treatments. There was no difference (P = 0.48) in the change in 12th-rib fat depth during the study; however, change in LM area was affected quadratically as the level of BM increased in the diet, with the greatest increase in LM area occurring in steers fed the 1x and 1.5x required arginine treatments. Balancing the AA:energy ratio did not affect G:F, DMI, or ADG; however, it increased deposition of lean in the LM quadratically. Level of dietary inclusion of BM as an RUP source does not affect its RUP value or efficacy of providing postruminal AA in growing steers.

Targeting fibroblast growth factor-inducible-14 signaling protects from chronic relapsing experimental autoimmune encephalomyelitis

The TNF-related weak inducer of apoptosis (TWEAK) is a TNF family member mediating proinflammatory effects by its receptor fibroblast growth factor-inducible-14 (Fn14). We studied the role of TWEAK/Fn14 in experimental autoimmune encephalomyelitis (EAE) by protein vaccination with TWEAK and Fn14 and recombinant TWEAK-DNA, respectively. TWEAK-DNA vaccination worsened the clinical course of EAE and increased central nervous system (CNS) inflammation. TWEAK increased the secretion of CCL2 [monocyte chemotactic protein-1 (MCP-1)] by CNS endothelial cells and astrocytes in vitro, suggesting CCL2 as a critical mediator of TWEAKs proinflammatory effects. Vaccination with the extracellular domain of TWEAK or with Fn14 resulted in the induction of specific inhibitory antibodies and an amelioration of EAE signs in two different models in rats and mice. Spinal cord inflammatory infiltrates were significantly diminished. Purified IgG from TWEAK- or Fn14-vaccinated rats prevented TWEAK-induced production of CCL2 by endothelial cells. Blocking Fn14 signaling represents a novel approach with potential for the treatment of CNS autoimmunity.

Cerebral blood flow and neurological outcome in the preterm infant

Cerebral blood flow (CBF) studies have provided some insight into pathophysiological mechanisms of cerebral damage in newborn children; their value in predicting brain damage, however, remains elusive. The purpose of our study was to evaluate the role of CBF measurements in predicting developmental outcome in preterm neonates at 18 months. Preterm babies with a gestational age of less than 34 weeks and a birth weight of less than 1500 g (n = 71) were enrolled in the study. CBF was measured by the noninvasive intravenous 133Xe method on three different occasions. We classified our measurements into three groups: depending on the time when performed group 1: between 2 and 36 h (n = 52); group 2: between 36 and 108 h (n = 44); group 3: between 108 and 240 h (n = 41). At the age of 18 months neurodevelopment testing was performed according to the Bayley mental and motor scales. Surviving infants had a higher mean CBF over the three groups than non surviving children (15.2 +/- 3.5 ml/100 g brain tissue/min vs 13.0 +/- 2.1 ml/100 g brain tissue/min, P < 0.05). There was no correlation of CBF with mental or motor development in our study population in either of the three groups.

CONCLUSION

In preterm infants basal CBF is higher in surviving than in non surviving infants, but there is no correlation of resting CBF and later neurological outcome.

Immunological detection and mutational analysis of the RNA2-encoded nematode transmission proteins of pea early browning virus

Pea early browning virus (PEBV) is transmitted between plants by root-feeding trichodorid nematodes. Mutagenesis studies have implicated two non-structural viral proteins in the transmission process. These two proteins [the 29 kDa ('29K') protein and the 23K protein] were expressed in bacteria and used to raise antibodies. In Western blotting experiments, the antibodies detected both of these virus proteins in leaves and roots of infected Nicotiana bethamiana and N. clevelandii plants. Periodate treatment of proteins transferred to nitrocellulose membranes suggested that the PEBV 23K protein may be glycosylated. A PEBV mutant was constructed lacking the complete 23K coding sequence. The mutant was able systemically to infect Nicotiana spp. but caused striking chlorotic ringspot leaf symptoms and stunting of both leaves and roots. These symptoms were absent in plants doubly-infected with the mutant and wild-type PEBV. The 23K gene deletion mutant was transmitted by nematodes at a much reduced frequency compared to wild-type virus, indicating that the 23K protein is involved in but not essential for vector transmission. Western immuno-blot and ELISA experiments revealed that the reduction in the nematode-transmissibility of PEBV carrying mutations in the 23K gene did not result from interference in the expression of the 29K transmission protein or from gross changes in the titre of virus in the roots of infected plants.

Replication of in vitro tobravirus recombinants shows that the specificity of template recognition is determined by 5' non-coding but not 3' non-coding sequences

Natural recombinant tobacco rattle tobravirus (TRV) isolates contain sequences from a different tobravirus, pea early browning virus (PEBV). To characterize the sequence requirements for viable recombinant formation hybrid cDNA clones of RNA2 of PEBV and TRV were assembled. Inclusion of 320 nt from the 5' terminus of PEBV or 335 nt from the 5' terminus of TRV in the hybrid RNAs was sufficient to permit their replication by, respectively, PEBV RNA1 or TRV RNA1 regardless of the origin of the 3' terminal region. However, PEBV RNA1 but not TRV RNA1 was sometimes able to support low level replication of RNA2 containing the heterologous 5' terminal region. In vitro translation of PEBV transcripts containing 5' noncoding region deletions supported the hypothesis that in vivo the PEBV coat protein (CP) is expressed from a subgenomic RNA and that, therefore, in the recombinants the CP subgenomic promoter probably is recognized by the replicase of the heterologous virus.

Cytogenetic evaluation of bone marrow cells from rats exposed to styrene vapor for one year

This report presents the cytogenetic findings in bone marrow cells of rats exposed to styrene vapor. Male and female Sprague-Dawley rats were exposed to 0,600 and 1000 ppm of styrene vapor by inhalation 6 hr per day, 5 days a week, for a period of one yr. Blind scoring of metaphase spreads prepared from bone marrow cells collected at the end of the last exposure revealed that neither the 600 ppm nor the 1000 ppm exposures to styrene vapor produced an incidence of chromosomal anomalies higher than those occurring spontaneously. It is interpreted that styrene is non-clastogenic within the present exposure regimen.