Non-melanoma skin cancer diagnosis: a comparison between dermoscopic and smartphone images by unified visual and sonification deep learning algorithms

Purpose

Non-melanoma skin cancer (NMSC) is the most frequent keratinocyte-origin skin tumor. It is confirmed that dermoscopy of NMSC confers a diagnostic advantage as compared to visual face-to-face assessment. COVID-19 restrictions diagnostics by telemedicine photos, which are analogous to visual inspection, displaced part of in-person visits. This study evaluated by a dual convolutional neural network (CNN) performance metrics in dermoscopic (DI) versus smartphone-captured images (SI) and tested if artificial intelligence narrows the proclaimed gap in diagnostic accuracy.

Methods

A CNN that receives a raw image and predicts malignancy, overlaid by a second independent CNN which processes a sonification (image-to-sound mapping) of the original image, were combined into a unified malignancy classifier. All images were histopathology-verified in a comparison between NMSC and benign skin lesions excised as suspected NMSCs. Study criteria outcomes were sensitivity and specificity for the unified output.

Results

Images acquired by DI (n = 132 NMSC, n = 33 benign) were compared to SI (n = 170 NMSC, n = 28 benign). DI and SI analysis metrics resulted in an area under the curve (AUC) of the receiver operator characteristic curve of 0.911 and 0.821, respectively. Accuracy was increased by DI (0.88; CI 81.9–92.4) as compared to SI (0.75; CI 68.1–80.6, p < 0.005). Sensitivity of DI was higher than SI (95.3%, CI 90.4–98.3 vs 75.3%, CI 68.1–81.6, p < 0.001), but not specificity (p = NS).

Conclusion

Telemedicine use of smartphone images might result in a substantial decrease in diagnostic performance as compared to dermoscopy, which needs to be considered by both healthcare providers and patients.

Skin cancer detection by deep learning and sound analysis algorithms: A prospective clinical study of an elementary dermoscope

Background

Skin cancer (SC), especially melanoma, is a growing public health burden. Experimental studies have indicated a potential diagnostic role for deep learning (DL) algorithms in identifying SC at varying sensitivities. Previously, it was demonstrated that diagnostics by dermoscopy are improved by applying an additional sonification (data to sound waves conversion) layer on DL algorithms. The aim of the study was to determine the impact of image quality on accuracy of diagnosis by sonification employing a rudimentary skin magnifier with polarized light (SMP).

Methods

Dermoscopy images acquired by SMP were processed by a first deep learning algorithm and sonified. Audio output was further analyzed by a different secondary DL. Study criteria outcomes of SMP were specificity and sensitivity, which were further processed by a F2-score, i.e. applying a twice extra weight to sensitivity over positive predictive values.

Findings

Patients (n = 73) fulfilling inclusion criteria were referred to biopsy. SMP analysis metrics resulted in a receiver operator characteristic curve AUC's of 0.814 (95% CI, 0.798–0.831). SMP achieved a F2-score sensitivity of 91.7%, specificity of 41.8% and positive predictive value of 57.3%. Diagnosing the same set of patients' lesions by an advanced dermoscope resulted in a F2-score sensitivity of 89.5%, specificity of 57.8% and a positive predictive value of 59.9% (P=NS).

Interpretation

DL processing of dermoscopic images followed by sonification results in an accurate diagnostic output for SMP, implying that the quality of the dermoscope is not the major factor influencing DL diagnosis of skin cancer. Present system might assist all healthcare providers as a feasible computer-assisted detection system.

Fund

Bostel Technologies.

Trial Registration clinicaltrials.gov Identifier: NCT03362138

Dermoscopy diagnosis of cancerous lesions utilizing dual deep learning algorithms via visual and audio (sonification) outputs: Laboratory and prospective observational studies

BACKGROUND

Early diagnosis of skin cancer lesions by dermoscopy, the gold standard in dermatological imaging, calls for a diagnostic upscale. The aim of the study was to improve the accuracy of dermoscopic skin cancer diagnosis through use of novel deep learning (DL) algorithms. An additional sonification-derived diagnostic layer was added to the visual classification to increase sensitivity.

METHODS

Two parallel studies were conducted: a laboratory retrospective study (LABS, n = 482 biopsies) and a non-interventional prospective observational study (OBS, n = 63 biopsies). A training data set of biopsy-verified reports, normal and cancerous skin lesions (n = 3954), were used to develop a DL classifier exploring visual features (System A). The outputs of the classifier were sonified, i.e. data conversion into sound (System B). Derived sound files were analyzed by a second machine learning classifier, either as raw audio (LABS, OBS) or following conversion into spectrograms (LABS) and by image analysis and human heuristics (OBS). The OBS criteria outcomes were System A specificity and System B sensitivity as raw sounds, spectrogram areas or heuristics.

FINDINGS

LABS employed dermoscopies, half benign half malignant, and compared the accuracy of Systems A and B. System A algorithm resulted in a ROC AUC of 0.976 (95% CI, 0.965-0.987). Secondary machine learning analysis of raw sound, FFT and Spectrogram ROC curves resulted in AUC's of 0.931 (95% CI 0.881-0.981), 0.90 (95% CI 0.838-0.963) and 0.988 (CI 95% 0.973-1.001), respectively. OBS analysis of raw sound dermoscopies by the secondary machine learning resulted in a ROC AUC of 0.819 (95% CI, 0.7956 to 0.8406). OBS image analysis of AUC for spectrograms displayed a ROC AUC of 0.808 (CI 95% 0.6945 To 0.9208). By applying a heuristic analysis of Systems A and B a sensitivity of 86% and specificity of 91% were derived in the clinical study.

INTERPRETATION

Adding a second stage of processing, which includes a deep learning algorithm of sonification and heuristic inspection with machine learning, significantly improves diagnostic accuracy. A combined two-stage system is expected to assist clinical decisions and de-escalate the current trend of over-diagnosis of skin cancer lesions as pathological. FUND: Bostel Technologies. Trial Registration clinicaltrials.gov Identifier: NCT03362138.

A hyperosmotic stimulus elevates intracellular calcium and inhibits proliferation of a human keratinocyte cell line

Occlusion has previously been used to treat psoriatic plaques and was shown to improve the condition. We investigated the consequences of applying a mechanical stress, in vitro, on the HaCaT keratinocyte cell line. A mechanical load applied to cells can be mimicked by a hyperosmotic stimulus. Exposure of HaCaT keratinocytes to different hyperosmotic solutions (final osmolarity in the range 350-600 mOsm, produced by sucrose addition) resulted in an inhibition of cell proliferation after 96 h of treatment. As keratinocyte maturation is regulated by calcium levels, we measured hyperosmotic-stimulus-induced changes of intracellular calcium ([Ca2+]i) by single-cell image analysis employing FURA-2/AM. The hyperosmotic stimulus produced a rapid transient 2.6-fold elevation of [Ca2+]i followed by a gradual decay to the basal level. The transients originated from extracellular as well as from intracellular calcium pools and did not respond to voltage-sensitive calcium channel blockers. The hyperosmotic stimulus was shown to increase the cellular expression of involucrin, a differentiation marker, following 72 h of incubation, as measured by flow cytometry. Treatment of cells with the [Ca2+]i chelator BAPTA/AM almost completely blocked the [Ca2+]i elevation, but did not alter cellular growth or the induction of differentiation observed after hyperosmotic stimulus. It is suggested that treatment of keratinocytes with hyperosmotic stimulus can induce short-time effects (calcium transients) as well as long-term cellular maturation.

Acoustic monitoring of intraoperative neuromuscular block

Standard methods for accurate intraoperative measurement of neuromuscular block are either expensive or inconvenient and are not used widely. We have evaluated a new method of monitoring neuromuscular block using a low-frequency microphone. The method is based on the phenomenon of low-frequency sound emission by contracting skeletal muscle. Acoustic monitoring (MIC) with an air-coupled microphone was used to evaluate intraoperative neuromuscular block in 25 anaesthetized patients. The MIC recorded the response of the adductor pollicis muscle to supramaximal electrical stimulation of the ulnar nerve with train-of-four stimuli. The ratios of the first response (TI) to control (TC) were used for evaluation. Data obtained from the MIC were compared with simultaneous recordings, from the same hand, of mechanomyography (FDT), electromyography (EMG) and accelerography (ACC). Throughout the operative procedure, TI/TC ratios of the acoustic method correlated with the three reference devices: FDT, 12 patients, 262 data sets, r = 0.86, bias (%MIC-%FDT) = mean -5.3 (SD 19.6)%; EMG, 18 patients, 490 data sets, r = 0.85, bias (%MIC-%EMG) = -0.39 (20.29)%; and ACC, 13 patients, 328 data sets, r = 0.91, bias (%MIC-%ACC) = -3.0 (15.6)%. We conclude that monitoring intraoperative neuromuscular block by a microphone which transduces low-frequency muscle sounds is clinically feasible.

A hyperosmotic stimulus regulates intracellular pH, calcium, and S-100 protein levels in avian chondrocytes

Cartilage is exposed to mechanical loads, generating at the level of single chondrocytes a hyperosmotic stimulus (HOS). The direct effect of HOS on second messenger pathways in avian chondrocytes was evaluated by fluorimetric and image analysis techniques. HOS caused an immediate intracellular acidification of 0.07 +/- 0.02 pH units (n = 7), followed by an initial pH recovery rate of 0.033 +/- 0.04 pH units/min towards the pre-stimulus baseline values. Concomitantly, the intracellular calcium ([Ca2+]i) responded with a transient rise from baseline value of 84.7 +/- 7.4 nM to peak level of 403.1 +/- 51.0 nM (n = 16, p < 0.001). The calcium response was abolished by two calmodulin inhibitors chlorpromazine and W-7. Since these inhibitors are known to be specific ligands of a S-100 protein, its intracellular staining was determined following HOS. The amount of immunodetectable S-100 protein was significantly increased following exposure to HOS (p < 0.05), and did not require an increase of [Ca2+]i. It appears that compression of cartilage is transduced into HOS of chondrocytes, and further elicits its effects through transient intracellular elevation of protons and calcium ions accompanied by increased staining of S-100 protein.

Hyperosmotic modulation of the cytosolic calcium concentration in a rat osteoblast-like cell line

1. The effects of hyperosmotic stress on cytosolic calcium concentration ([Ca2+]i) were studied by ratio image analysis in single cells of an osteoblast-like bone cell line (RCJ 1.20) loaded with fura-2 AM. 2. The ratio (340 nm/380 nm) of steady-state [Ca2+]i in resting osteoblasts kept in Hepes-buffered medium was 0.82 +/- 0.04. A hyperosmotic stimulus (200 mosmol l-1 sucrose) produced a [Ca2+]i transient with a peak ratio of 1.28 +/- 0.09, which decayed with an apparent half-life (t1/2) of 42.7 +/- 2.6 s. 3. The hyperosmotically induced [Ca2+]i transients were insensitive to verapamil, diltiazem or nifedipine, which excludes the involvement of dihydropyridine-sensitive Ca2+ channels in the process. Non-specific Ca2+ channel blockers (Mn2+, Ni2+, La3+ or Gd3+) partially abolished the hyperosmotically induced [Ca2+]i elevation, indicating the contribution of extracellular Ca2+ influx. 4. A hyperosmotic stimulus applied in Ca(2+)-free medium (0.5 mM EGTA) lowered the [Ca2+]i peak to a ratio of 0.96 +/- 0.08 (P < 0.001) compared with a Ca(2+)-containing medium. This suggests that the [Ca2+]i increase is due to extracellular influx, as well as release from an intracellular Ca2+ pool. 5. Application of thapsigargin (0.5 microM), a specific inhibitor of endoplasmic reticulum Ca(2+)-ATPase, in Ca(2+)-free medium caused transient [Ca2+]i elevation to peak ratios of 1.33 +/- 0.09, and completely abolished the [Ca2+]i response to a hyperosmotic stimulus. This implies the existence of a thapsigargin-sensitive intracellular pool of Ca2+ that is mobilized by hyperosmotic stimulus.(ABSTRACT TRUNCATED AT 250 WORDS)

Effects of radiologic contrast media on human endothelial and kidney cell lines: intracellular pH and cytotoxicity

RATIONALE AND OBJECTIVES

Radiologic contrast media (CM) are hyperosmotic compounds injected undiluted into a patient's blood, in which they contact endothelial cells. For some types of cultured cells, the application of a hyperosmotic stimulus may cause intracellular pH (pHi) acidification that is related to the extent of hyperosmolality and that ultimately influences cellular function. Accordingly, endothelial and kidney cells, two types of cells known to be exposed to CM effects, were treated at relevant iodine concentrations with various CM (320-1500 mOsm) to determine whether cell exposure to CM can disturb the pH(i) and to examine the contribution of CM to cellular cytotoxicity.

METHODS

Ionic (n = 3) and nonionic (n = 3) CM were compared. Changes in the pH(i) of human vascular endothelial and kidney cell lines were monitored by use of a pH-sensitive fluorescent dye (2',7'-bis(carboxyethyl)-5(6)-carboxyfluorescein acetoxymethyl ester). The viability of cells treated with CM was determined by measuring the reduction of a tetrazolium salt (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenlytetrazolium bromide) to violet formazan, a reaction that requires the activity of mitochondrial dehydrogenase; this measurement was made with a microplate reader.

RESULTS

The pH(i) of endothelial and kidney cells exposed to 40-60 mg of iodine per milliliter of CM showed acidification (approximately 0.2 pH unit). Within minutes, gradual pH(i) alkalinization to baseline values occurred. The return to baseline values was slower with ionic compounds than with nonionic CM (P < 0.001). Nonionic agents caused less cellular damage than did ionic CM.

CONCLUSIONS

The pH(i) is involved in the immediate intracellular transduction of CM effects in vitro. The exposure of cells to ionic CM is more detrimental than is exposure to nonionic CM, as demonstrated by disturbances in the cytosolic pH and by long-term effects on cell viability.

The control of intracellular pH in cultured avian chondrocytes

1. Mechanical loading of cartilaginous tissue generates an increase in the concentration of cations in the extracellular matrix. This includes a decrease of the extracellular pH (pHo), which is known to affect the intracellular pH (pHi), thereby modifying the intracellular metabolism. Thus, the regulation of pHi is essential for the physiological function of cartilage. The fluorescent pH-sensitive dye 2',7'-bis(carboxyethyl)-5(6)-carboxyfluorescein acetoxymethyl ester (BCECF AM) was employed in order to assess the mechanisms responsible for control of the pHi in an embryonic avian chondrocyte cell suspension. 2. Steady-state pHi in the absence of physiological HCO3- was 7.15 +/- 0.01 pH units as compared to a pHi of 6.94 +/- 0.02 pH units in its presence (P < 0.01). The intrinsic buffering power of chondrocytes (beta i) was 38.9 mM/pH unit and the total buffering capacity (beta T) was 65.8 mM/pH unit. 3. Cells maintained in a Hepes-buffered solution were exposed to an intracellular acid load by the NH4+ prepulse technique (20 mM NH4Cl). The initial rate of pHi recovery was 0.106 pH units/min (n = 18). Amiloride (0.33 mM), an inhibitor of the Na(+)-H+ exchanger, or replacement of external sodium [Na+]o with choline induced a 60% inhibition of the recovery rate, indicating a predominant involvement of this antiporter in the response to intracellular acidification. 4. H(+)-ATPase inhibitors (oligomycin 20 micrograms/ml; N,N;-dicyclohexylcarbodiimide (DCC), 0.5 mM; N-ethylmaleimide (NEM), 0.25 mM) and iodomycin (2 mM), a metabolic cell suppressor, reduced acid extrusion by 25% as measured by the NH4Cl prepulse in Hepes-bathed cells. 5. Chondrocytes transferred from a Hepes-buffered solution to a 5% CO2-25 mM HCO3- medium (HCO3- solution) underwent a pHi decrease of approximately 0.20 pH units, followed by a regulatory alkalinizing response of 0.118 pH units/min. The Na(+)-H+ exchanger was responsible for only 15% of this alkalinization (amiloride, 0.33 mM), in contrast to its primary role in HCO(3-)-free solution. 6. The activity of a Na(+)-dependent Cl(-)-HCO3- exchanger in physiological HCO3- solution was estimated by addition of the inhibitors 4-acetamido-4'-isothiocyanatostilbene-2,2'-disulphonic acid (SITS; 0.5 mM) or diisothiocyanatostilbene-2,2'-disulphonic acid (DIDS; 100 microM) and by the suspensions of chondrocytes in a Na(+)-free solution. Acidification performed under these conditions resulted in a 45% inhibition of the recovery rate as compared to control rates.(ABSTRACT TRUNCATED AT 400 WORDS)

Hyperosmotic activation of the Na(+)-H+ exchanger in a rat bone cell line: temperature dependence and activation pathways

1. The hyperosmotic activation of the Na(+)-H+ exchanger was studied in an osteoblast-like rat cell line (RCJ 1.20). The activation was monitored by recording the intracellular pH (pHi) changes employing double excitation of the pH-sensitive fluorescent dye 2'7'-bis(carboxyethyl)-5(6)-carboxyfluorescein acetoxymethyl ester (BCECF-AM). 2. Exposure of the cells to a hyperosmotic HCO(3-)-free medium at 37 degrees C produced an initial cytosolic acidification of 0.05 pH units followed by a lag period and an alkalinization overshoot of about 0.2 pH units, without a concomitant change of the free cytosolic calcium [Ca2+]i by the use of Fura-2 calcium-sensitive probes. This response was completely inhibited by amiloride (0.33 mM) or by Na+ depletion from the external medium and insensitive to the extracellular Cl- replacement, indicating the involvement of a Na(+)-H+ exchanger in the hyperosmotic response. 3. Hyperosmotic stimuli (200 moSM sucrose) applied in the temperature range of 17-37 degrees C demonstrated a shortening of the lag period preceding alkalinization and an increased rate of proton extrusion upon temperature elevation. The biochemical reaction underlying the lag period and the proton extrusion resulted in apparent activation energies of 19 and 29 kcal mol-1, respectively, as calculated from the appropriate Arrhenius plots. 4. Stimulation of the exchanger under isosmotic conditions by 25 nM 4 beta-phorbol 12-myristate 13-acetate (PMA) and 0.1 mM vanadate resulted in an amiloride-sensitive pHi increase of about 0.08 pH units. The hyperosmotic stress was additive to the stimulatory effects of these agents, suggesting an independent hyperosmotic activation pathway. 5. The hyperosmotic activation of the Na(+)-H+ exchanger was independent of cAMP, cGMP, cytosolic Ca2+ and protein kinase C. Thus, none of the classical transduction mechanisms seem to be involved directly in the hyperosmotic activation of the antiporter. 6. The pHi response induced by the hyperosmotic stress was abolished by two calmodulin inhibitors, W-7 and chlorpromazine (50% inhibition, Ki at 28 and 20 microM, respectively), 20 microM cytochalasin B, but not by 10 microM colchicine. The results suggest the involvement of actin and calmodulin-like structural elements of the cytoskeleton in the transduction process leading to the activation of the Na(+)-H+ exchanger.

Regulation of the Na+/H+ exchanger under conditions of abolished proton gradient: isosmotic and hyperosmotic stimulation

Activation of the Na+/H+ exchanger following isosmotic and hyperosmotic stimuli was investigated in an osteoblast cell line (RCJ 1.20). The pH dependence of the transporter activity was studied under conditions of abolished proton gradient (pHi = pHo) across the membrane. The isotonic response is Na+o dependent, increases towards higher pH-values, displaying a sigmoidal dependence on pHi = o (Hill coefficient approximately 1.8) and is controlled by pHo. The greater than first order dependence on pH suggests that H+o inhibits the exchange beyond the rate expected from competition with the Na+o alone. This may be due to the existence of an external H+ regulatory site with a negative cooperative effect on the intra- or extracellular transport site. The hyperosmotic activation is Na+o independent, parallels the sigmoidal pH dependence of the isosmotic stimulus (Hill coefficient approximately 2.0) and is mediated through an increase of the Vmax without a change in the intracellular proton sensitivity.