Sage Journals: Discover world-class research

Abstract

Objectives:

To evaluate the performance of noninvasive, elastic scattering spectroscopy, algorithm-powered device (DermaSensor) to detect melanoma and basal and squamous cell cancers in the primary care setting.

Patients & Methods:

DERM-SUCCESS, a blinded, prospective, multicenter pivotal study, enrolled adult patients between August 17, 2020, and December 9, 2021, with lesions that their primary care physicians (PCPs) suspected of skin cancer at clinics in the US (n = 18) and Australia (n = 4). These lesions were assessed by PCPs and scanned with the DermaSensor device. Biopsy specimens were collected, and histopathologic analysis was performed by dermatopathologists. The diagnostic performance of the device, dermatopathologist discordance, and subgroup analyses of clinical interest were calculated.

Results:

Of the 1579 skin lesions enrolled, dermatopathologic analysis identified 224 (14.2%) cancers. Device sensitivity was 95.5% (95% CI, 91.7%-97.6%) overall and 96.3% (92.9%-98.4%) for patients in the FDA-approved age group 40 years and older (90.2% for melanoma, 97.8% for basal cell carcinoma, and 97.7% for squamous cell carcinoma). Device specificity was 20.7%. The negative predictive value was 96.6%, and the positive predictive value was 16.6% (NNB 6). The device misclassified as “monitor” rather than “investigate further” 4 keratinocyte carcinomas and 4 melanomas in patients aged 40 years or older (n = 8, 0.5% of lesions, 3.7% of cancers biopsied).

Conclusions:

The DermaSensor device is an easy-to-use, point-of-care, hand-held skin cancer adjunctive diagnostic device with high sensitivity and NPV to help inform PCP decision-making about skin lesions suspicious for cancer that need further evaluation and those that may be monitored.

Keywords

basal cell carcinoma DermaSensor elastic scattering spectroscopy melanoma primary care skin cancer squamous cell carcinoma

Introduction

The incidence of basal cell carcinoma (BCC), squamous cell carcinoma (SCC) (collectively, keratinocyte carcinomas [KC]), and malignant melanoma is increasing worldwide. Skin cancer will be diagnosed in 1 of 5 Americans during their lifetime, making it the most common type of cancer in the US.¹ In 2022, an estimated 5.4 million cases of KC were diagnosed in 3.3 million people, and an estimated 2500 to 15 000 patients died of SCC.^1
-4 An estimated 99 780 new cases of invasive melanoma and 97 920 cases of melanoma in situ were diagnosed in 2022, and 7650 people died of melanoma.¹ Most skin cancer is curable if diagnosed early, including KC, melanoma in situ, and nonulcerated malignant melanoma with a Breslow depth less than 1 mm.¹

Skin biopsy and subsequent histopathologic analysis is the standard for skin cancer diagnoses.⁵ The number needed to biopsy (NNB) ratio of total biopsies to identify 1 histopathologic-confirmed skin cancer is commonly used to compare the diagnostic accuracy of providers who perform skin biopsy and is the inverse of positive predictive value (PPV).⁶ Training and experience in dermatologic care correlates inversely with NNB and directly with diagnostic specificity and PPV.^6
-8 One study at a large academic medical center found an NNB to find 1 melanoma of 14.33 for dermatologists, 27.80 for PCPs, and 53.56 for other primary care providers.⁸ And, PCP sensitivity for correctly managing (i.e., refer or biopsy) skin cancer by naked eye examination is well below that of dermatologists, ranging from 54% to 88%.^9,10

With only 1 dermatologist for every 30 000 people in the US, early skin cancer detection cannot depend on dermatologists. Improved PCP performance is necessary to improve detection and avoid unnecessary biopsies and referrals of benign lesions for biopsy.¹¹ Dermoscopy training greatly improves PCP sensitivity and specificity;^10,12
-17 however, the requirements for additional training and ongoing practice to maintain dermoscopy competency has limited its use in the US to only 8% of PCPs.^18,19 Despite availability of a range of education and algorithm interventions to improve PCP diagnostic ability, outcomes are inconsistent, difficult to replicate, and challenging to maintain.²⁰

The Nevisense system (Scibase) is an electrical impedance spectroscopy device approved by the US Food and Drug Administration (FDA) for melanoma detection (sensitivity, 96.6%; specificity, 34.4%). Its use is restricted to dermatologists and only lesions concerning for melanoma.²¹ Because seborrheic keratoses generate high electrical impedance scores, they must be excluded via prescreening.

Artificial intelligence-assisted smartphone applications that assess clinical and dermoscopic images of skin lesions are becoming increasingly available though none are authorized for use by the FDA.^22,23 Although such image-based software may eventually achieve dermatologist-level sensitivity and specificity,^23
-26 current performance (early 2025) is below standard of care,^27,28 increasing use of health care resources without added benefits.^29,30 These applications are unable to evaluate cellular characteristics like ESS.

An affordable, automated point-of-care device is needed to improve PCP skin cancer detection and reduce unnecessary referrals for biopsy. Elastic scattering spectroscopy (ESS) is a noninvasive method for obtaining a spectral recording of photons scattered by chromophores in tissues, facilitating detection of cellular elements indicative of malignant neoplasms in skin³¹ and other tissues.^32
-36

Methods

Trial Design

The DermaSensor Study of Primary Care Physician Use of Elastic Scattering Spectroscopy (DERM-SUCCESS) was a prospective, blinded validation trial designed in collaboration with the FDA for clearance of a novel ESS, algorithm-powered device (DermaSensor device, manufactured by DermaSensor, Inc.) for evaluating lesions suggestive of skin cancer. The primary aim of the DERM-SUCCESS trial was to measure the performance of the device for lesions suggestive of melanoma, BCC, and/or SCC. The secondary aim was to compare device sensitivity to a performance goal of 90%, which is based on the published dermatologist sensitivity range of 81% to 96%.^37
-40

This study was reviewed and approved by the 3 Institutional Review Boards, each overseeing separate sites. Written informed consent was obtained for all participants before enrollment. Study participants received a small stipend.

Setting

Participants were enrolled in a convenience sample of suburban and rural family medicine and internal medicine practices, 18 in the US and 4 in Australia, of experienced PCPs who conduct a high volume of skin exams and perform their own skin biopsies from August 17, 2020, through December 9, 2021. Nine of the 22 study sites were in towns with populations under 15 000 people.

Participants

All patients seeing study PCPs during study enrollment with a suspicious lesion who met eligibility criteria, and were willing to participate were enrolled (Figure 1). Eligibility criteria included patients aged 22 years or older with a skin lesion(s) suggestive of melanoma, BCC, and/or SCC (based on visual inspection by PCP) requiring biopsy. Excluded lesions included those less than 2.5 mm or greater than 15 mm in diameter (smaller lesions were excluded due to risk of the 2.5 mm probe tip placement on perilesional skin; larger lesions were excluded since the 5 recordings sample a limited volume of the lesional tissue); located on mucosal or acral skin, within 1 cm of the eye, at a previous biopsy or surgical site, or in acute sunburned skin; completely covered with thick crust, erosion, or ulceration (without a nonulcerated area ≥2.5 mm) or containing foreign matter. Dermoscopy was not used for lesion assessment. After target numbers of KCs were reached (see protocol below), participating PCPs were instructed to continue to enroll patients with pigmented lesions until the number of melanomas required by the FDA for this trial were enrolled. Participant sex, age, ethnicity, and self-reported race were recorded, and Fitzpatrick skin type was assessed by the patient and recorded by PCPs.

Figure 1.

Flow diagram of included and excluded study data. The ITT population comprised all patients and lesions that were included in the safety analysis, and the mITT population comprised all patients and lesions that were included in the effectiveness analysis.

Device

The battery-operated, handheld DermaSensor device (Figure 2) emits pulses of light (200 microseconds/pulse) that span wavelengths from near-ultraviolet, through visible, to near-infrared. Two optical fibers are enclosed in a biocompatible 2.5-mm sterilizable fiberoptic tip that is placed in contact with the lesion surface. The volume of tissue evaluated with the device for each spectral recording was estimated computationally with Monte Carlo simulations to be approximately 0.7 × 0.4 × 0.5 mm (length × width × depth). Five spectral scans at various points on the lesion can be performed in about 10 s total time per lesion to record light reflectance of tissue structure and architecture (e.g., nuclear and chromatin characteristics) with the algorithm, then comparing those spectral readings to lesions in the training data set.

Figure 2.

Elastic scattering spectroscopy device for analysis of suspected skin cancer lesions. An example patient number (#234-1234), with a negative monitor or positive investigate further result, is shown. To investigate further results, a representative spectral score and corresponding definition are shown. (© 2023 DermaSensor, Inc; used with permission.).

The device AI developed algorithm was trained on more than 10 000 spectral scans and 2000 lesions. In accordance with requirements by the FDA for a pivotal study, the algorithm was developed before study enrollment. The algorithm was upgraded and then used for analysis of the blinded dataset based on data from 3 other clinical trials before study completion. The device output to which investigators and dermatopathologists were blinded is either monitor or investigate further, providing a spectral score of 1 through 10 for the latter (see Figure 2). A higher spectral score indicates greater spectral similarity with that of malignant lesions.

Study Sequence

The PCPs electronically recorded a lesion(s) description. The lesion(s) was then scanned with the ESS device. Both PCPs and patients were blinded to the results. A biopsy specimen was collected according to the PCP standard of care and provided to an independent dermatopathologic core laboratory for histopathologic analysis. As with the PCPs, the dermatopathologists were blinded to device output.

All specimens were analyzed by 2 dermatopathologists who had to independently reach diagnostic consensus using MPATH-Dx criteria for melanocytic lesions⁴¹ Discordance (benign or malignant) between the first 2 dermatopathologists was adjudicated by a third dermatopathologist and by 2 additional dermatopathologists for lesions diagnosed by at least 2 of the initial dermatopathologists as highly atypical, with differing diagnoses (malignant, benign, or atypical) by each of the initial 3 dermatopathologists, and/or diagnosed as malignant by 1 dermatopathologist but not as malignant by the adjudicator. If the 2 additional dermatopathologists disagreed, the lesion was excluded from analysis. Patients’ participation in the study was complete after their enrolled lesion(s) was biopsied. All biopsy results were communicated to patients by their PCP and standard of care followed.

Data Analysis

Device sensitivity was tested with an α level of .025 and compared by using a method-of-moments approach for clustered matched-pair data⁴¹ to account for possible within-patient correlation for patients with multiple lesions. The device’s ability to detect malignant lesions better than random chance (sensitivity + specificity > 1) was tested with generalized estimating equations logistic regression with a noninferiority margin of 10% by using a 1-sided exact binomial test with an α level of .025, with each patient treated as a random effect and assuming an exchangeable within-patient correlation structure to account for multiple within-patient lesions. The secondary hypothesis was tested for performance in comparison with a 90% threshold. Level of agreement between dermatopathologists was assessed with the Cohen κ statistic. For all other statistical inferences, 2-sided Wilson or exact confidence intervals were determined as appropriate, and unless specified, P < .05 was considered statistically significant. All statistical analyses were performed with SAS software, v9.4 (SAS Institute Inc.).

Results

A total of 1028 patients (1598 lesions) consented to participate in the study (Figure 1), of whom 7 did not meet the inclusion criteria and 16 (19 lesions) were excluded because their lesions did not meet eligibility criteria, device deficiencies, or a dermatopathologic consensus was not reached. The resulting 1005 patients (1579 lesions, 1451 in the US and 128 in Australia) were included in our analyses.

Slightly more than half of participants were women (51.5%), and the majority (97.1%) were White (Table 1). Mean (SD) participant age was 58.5 (15.1) years, and 138 patients (13.7%) were younger than 40 years. Most participants (72.5%) had Fitzpatrick skin types I through III, with most (35.0%) having type III skin. Most patients had 1 (65.4%) or 2 (20.6%) eligible lesions analyzed in the study. Of 1579 lesions, 998 (63.2%) had darker pigmentation, and 581 (36.8%) had lighter pigmentation. Nearly two-thirds (60.6%) of lesions were elevated or papular, and 67.8% were discovered by the patient. Mean (SD) lesion size was 6.7 (2.9) mm in length and 5.2 (2.2) mm in width. All biopsy specimens collected were clinically suspicious for a malignant neoplasm. Histopathologic analysis by the dermatopathologists indicated that 1355 lesions were benign. Of 224 malignant lesions, 176 were KCs (90 BCC and 86 SCC) and 48 were melanocytic high-risk lesions (19 severely atypical nevi/atypical junctional melanocytic proliferations and 29 invasive and in situ melanomas) (Table 2). Of the 48 melanocytic lesions, 7 were on patients younger than 40 years. No KCs were identified for patients younger than 40 years. Severely atypical nevi, atypical junctional melanocytic, and melanoma in situ lesions were grouped in the analysis as high-risk melanocytic lesions with melanomas by FDA requirement to ensure referral to dermatology due to the close histopathological relationship and lack of reproducibility of diagnostic differentiation between them on histopathology.^41,42

Table 1.

Patient (n = 1005) and Lesion (n = 1579) Characteristics.

Characteristic	Value^a
Patient characteristics
Sex
Men	487 (48.5)
Women	518 (51.5)
Age, y	58.5 (15.1)
Race
Asian	9 (0.9)
Black or African American	7 (0.7)
Native Hawaiian or Other Pacific Islander	3 (0.3)
White	976 (97.1)
Other/multiracial	10 (1.0)
Fitzpatrick skin type
I (always burns, never tans)	99 (9.9)
II (always burns, tans minimally)	278 (27.7)
III (sometimes mild burn, tans uniformly)	352 (35.0)
IV (burns minimally, always tans well)	148 (14.7)
V (very rarely burns, tans very easily)	110 (10.9)
VI (never burns)	18 (1.8)
Risk factors
Weakened immune system	32 (3.2)
Family history of skin cancer	332 (33.0)
Fair skin, freckling, light hair	362 (36.0)
Many moles and/or dysplastic nevi	331 (32.9)
Number of eligible lesions
1	657 (65.4)
2	207 (20.6)
3	78 (7.8)
4	41 (4.1)
5	22 (2.2)
Lesion characteristics	(n=1,579)
Anatomic location
Head	247 (15.6)
Arm	299 (18.9)
Leg	207 (13.1)
Trunk	826 (52.3)
Length, mm	6.7 (2.9)
Width, mm	5.2 (2.2)
Elevation
Flat	622 (39.4)
Elevated	957 (60.6)
Surface texture
Smooth	864 (54.7)
Rough	715 (45.3)
Pigmentation
Lighter	581 (36.8)
Darker	998 (63.2)
Discovery
Patient	1,070 (67.8)
Family member/partner	79 (5.0)
Health care clinician	430 (27.2)

Patient age, lesion length, and lesion width summarized as mean (SD), and all other variables summarized as No. (%) of patients or lesions.

Table 2.

Dermatopathologic Risk Classifications and Diagnoses (n = 1579).

Dermatopathologic classification or diagnosis	No. (%)
Malignant lesions	224 (14.2)
Melanoma^a	48 (3.0)
Basal cell carcinoma	90 (5.7)
Squamous cell carcinoma	86 (5.4)
Benign lesions	1,355 (85.8)
Benign melanocytic nevus	500 (31.7)
Seborrheic keratosis	490 (31.0)
Actinic keratosis	71 (4.5)
Lentigo	65 (4.1)
Other	229 (14.5)

Includes 29 melanomas (invasive and in situ) and 19 highly atypical lesions (atypical junctional melanocytic proliferations and severely atypical nevi).

The level of agreement between the 2 core laboratory dermatopathologists was 81.3% (Cohen κ, –0.1; 95% CI, –0.2 to 0.0); they agreed that 39 of the 48 melanomas were melanoma (or its near-equivalent, e.g., severe atypia). One diagnosed 44 of 48 (91.7%) melanomas; the other diagnosed 43 of 48 (89.6%) melanomas. For patients 40 years and older, the sensitivity of the 2 dermatopathologists was 90.2% (37 of 41 melanomas). Thus, the device sensitivity of 90.2% for melanoma was identical to that of both dermatopathologists (i.e., all 3 correctly diagnosed 37 of 41 melanomas). Five lesions excluded from primary analysis because of a lack of consensus had various diagnoses on the spectrum from compound melanocytic nevi to severely atypical nevi. Exclusion of these lesions was documented before study database lock and unblinding of the study results.

Overall sensitivity of the ESS device for detecting skin cancer was 95.5% (95% CI, 91.7%-97.6%) ( Table 3 ). For the FDA-approved target group of patients aged 40 years and older, the device sensitivity for skin cancer was 96.3% (95% CI, 92.9%-98.4%), including 90.2% (95% CI, 76.9%-97.3%) for melanoma, 97.8% (92.2-99.7%) for BCC, and 97.7% (91.9%-99.7%) for SCC. For all ages, device sensitivity was 87.5% (95% CI, 76.4%-93.8%) for melanoma. Patient-level sensitivity, defined as the proportion of participants with skin cancer for whom the device correctly detected at least 1 malignant lesion was 97.4% (95% CI, 94.2%-99.2%) for all skin cancers and 95.1% (95% CI, 83.5%-99.4%) for melanoma for patients aged 40 years and older (see Table 3). For patients aged 40 years and older, patient-level sensitivity was 97.4% (95% CI, 94.2%-99.2%) for all skin cancers and 95.1% (95% CI, 83.5%-99.4%) for melanoma for patients aged 40 years and older. When compared with the performance goal of 90%, device sensitivity was both noninferior (P < .001) and superior (P = .002). The device also had significantly higher odds (odds ratio, 4.93; 95% CI, 2.84-8.56; P < .001) of detecting malignant lesions than random chance.

Table 3.

Sensitivity of Device Using Dermatopathologic Diagnoses as Standard

Diagnosis	Sensitivity, % (95% CI)^a
All skin cancers	95.5 (91.7-97.6)
All skin cancers (patient level)^b	97.0 (93.6-98.9)
All skin cancers (patients ≥40 years)	96.3
Melanoma^c	87.5 (76.4-93.8)
Melanoma (patient level)^b	93.6 (82.5-98.7)
Melanoma (patients ≥40 years)	90.2 (76.9-97.3)
Melanoma (patients ≥40 years, patient level)^b	95.1 (83.5-99.4)
BCC^d	97.8 (91.3-99.5)
BCC (patient level)^b	97.4 (90.9-99.7)
SCC^d	97.7 (91.1-99.4)
SCC (patient level)^b	98.7 (93.1-100.0)

Abbreviations: BCC, basal cell carcinoma; SCC, squamous cell carcinoma.

Values for 95% CI calculated accounting for within-subject correlations by using the Wilson method described by Saha et al.⁴³

Patients with at least 1 positive device result were coded as device positive at the patient level.

For the device’s indicated use for patients aged 40 years and older, the device sensitivity was 90.2%.

For BCC and SCC, no malignant lesions were recorded for patients younger than 40 years; therefore, overall and patient-level sensitivity were unchanged for the indicated use population.

Monitor (i.e., negative) device results did not occur for any melanomas that exceeded 0.7 mm in depth (i.e., stage IA, T1a). The device misclassified 1 severely atypical nevus, 1 melanoma in situ, and 4 melanomas with 0.7 mm or less thickness although higher spectral scores provided by the device directly correlated with increased malignancy nearly reaching an investigate further (i.e., positive) result (See Table 4). Overall specificity of the ESS device was 20.7% (95% CI, 18.5%-23.1%) for correctly classifying biopsied lesions as benign, and 20.3% (95% CI, 18.0%-22.7%) for patients aged 40 and above.

Table 4.

Clinical Characteristics and Histopathologic Findings for Missed Melanomas.

Clinical characteristics	Device result	Algorithm output^a	Histopathologic findings	Body location
Flat, smooth, dark, 4 × 4 mm	Monitor	0.305	Lentiginous compound melanocytic nevus with highly atypical characteristics	Upper back
Elevated, smooth, dark, 5 × 4 mm	Monitor	0.418	Melanoma in situ	Upper back
Flat, smooth, dark, 10 × 6 mm	Monitor	0.392	0.4-mm thickness melanoma	Forearm
Flat, smooth, dark, 4 × 4 mm	Monitor	0.412	0.4-mm thickness melanoma	Shoulder
Flat, smooth, dark, 10 × 8 mm	Monitor	0.442	0.6-mm thickness melanoma	Face, cheek
Elevated, rough, dark, 7 × 7 mm	Monitor	0.449	0.7-mm thickness melanoma	Upper leg

The device algorithm internally generates a numeric value based on a spectral score between 0 and 1, which is then converted to the device output of either a negative result of monitor (i.e., 0) or positive result of investigate further with a corresponding confidence score (1-10). The algorithm cutpoint for a investigate further result is a numeric value of 0.450; thus, 4 melanomas were within 0.032 of being a true positive result, and the highly atypical lesion was the only lesion missed by a large margin.

Although the dataset included fewer patients with Fitzpatrick skin type IV through VI (27.5%), no meaningful differences in sensitivity and specificity were evident between patients with Fitzpatrick skin type I through III (sensitivity, 96.5%; 95% CI, 92.6%-98.7%; specificity, 18.7%; 95% CI, 16.2%-21.5%) and IV through VI (sensitivity, 92.2%; 95% CI, 81.1%-97.8%; specificity, 25.1%; 95% CI, 20.9%-29.7%).

The AUROC curve for the device was calculated as 0.7796 which would be classified as very good for diagnostic devices.⁴⁴

The overall NPV of the device was 96.6% (95% CI, 93.5%-98.2%), which indicates that a lesion with a device result of monitor had a 3.4% chance of being malignant. Overall PPV of the device for skin cancer detection was 16.6% (95% CI, 14.2%-19.3%), which indicates that a lesion with a device result of investigate further had a 16.6% chance of being malignant (i.e., NNB of 6.0). For investigate further results, device PPV increased with increasing spectral score. Of further results, 37% of lesions had a spectral score of 1-3, which correlated with a 5.9% probability of malignancy, increasing to a PPV of 49.6% PPV with spectral scores of 9 to 10 (See Figure 3).

Figure 3.

Positive Predictive Value (PPV) According to Spectral Score for Investigate Further Results. Device PPV ranged from 6.4% (95% CI, 3.7%-10.9%) for the lowest spectral score of 1 to 61.3% (95% CI, 48.3%-72.9%) for the highest score of 10. The device PPV was 7.4% (95% CI, 5.5%-9.9%; NNB, 13.5) for low (1-5) spectral scores and 33.2% (95% CI, 28.8%-37.9%; NNB, 3.0) for high (6-10) scores (P < .001). When spectral scores were further categorized as low (1-3), medium (4-7), or high (8-10), the PPV was 5.9% (95% CI, 4.1%-8.5%; NNB, 16.9) for low scores, 18.4% (95% CI, 14.8%-22.7%; NNB, 5.4) for medium scores, and 39.6% (95% CI, 33.4%-46.2%; NNB, 2.5) for high scores (P < .001).

Discussion

The DermaSensor device’s 96.3% sensitivity for detecting skin cancer and 96.6% NPV classifies it a very good diagnostic device for risk stratifying suspicious skin lesions. The device was approved by the FDA in January 2024 based on this data for skin cancer risk assessment for patients 40 years and older, making it the first such device available for use by PCPs.²³

Since spectral scores are directly related to increasing malignancy risk, the device may find good utility for prioritizing timing of further care (e.g., urgency of referral to dermatology). The NNB for all skin cancers in this data set skewed towards a high percentage of melanocytic lesions was 7.1 for PCPs and 6.0 for the device. If PCPs had been unblinded and device output used to inform biopsy decisions, they would have decreased their biopsy frequency in the approved 40 years and older population by 18.4% (1288 vs 1579 biopsies) and would have monitored rather than investigated further 3.7% of cancers (4 KCs and 4 melanomas or 0.5% of lesions). For the 37% of investigate further lesions with a low spectral score of 1 to 3 which represented a 5.9% likelihood of malignancy, PCPs may have further decreased unnecessary care by monitoring lesions that were also of low clinical concern (e.g., lesions not suggestive of melanoma and/or patients with limited risk factors).

The device correctly classified 90.2% of melanomas as investigate further for patients 40 years and older (87.5% of melanomas overall)—an accuracy very similar to the individual dermatopathologists when compared to dermatopathology consensus. Discordance for melanoma diagnoses between dermatopathologists is common as observed in a trial of 187 US dermatopathologists randomly assigned to interpret the same lesions months apart where intraobserver and interobserver concordance was under 50% for diagnoses ranging from moderately dysplastic nevi to early-invasive melanoma, though higher for lesions at the ends of the spectrum—benign nevi and invasive melanoma.⁴²

Patient-level device sensitivity was 95.1% for the device to read investigate further for at least 1 lesion in any given patient with melanoma. Two recently published studies of the same device and algorithm in dermatology practices reported similar high device sensitivity for both KC and melanoma and a NNB congruent with that of dermatologists.^45,46 Another recently published small study using the same device and algorithm on lesions of concern to patients in 1 primary care practice found the device sensitivity to be 90%, but a specificity for correctly classifying benign lesions of concern to patients of 60.7% overall and 76.9% for pigmented benign lesions.⁴⁷ This study will need to be repeated in additional practices; but the device may be able to decrease unnecessary referrals for biopsy and prioritize referrals.

By analyzing ESS data from suspicious lesions, the device provides immediate risk information to the PCP to inform lesion management. This may improve referrals by PCPs and enhance referral adherence by patients. Pigmented lesions of concern for melanoma with a monitor result should still be monitored for changes, as should all concerning skin lesions. Future studies will help clarify re-examination frequency and lesion selection criteria for device use and device impact on unnecessary referral rates from PCP practices.

Limitations

Our study was limited by the exclusion criteria (e.g., fully crusted lesions which may have excluded some squamous cell cancers and very small and large lesions) and included only lesions identified as concerning and warranting biopsy by PCPs who frequently perform biopsies for skin cancer and who were blinded to device results. Enrollment of only lesions suspicious for melanoma after achieving the target sample size for BCC and SCC resulted in a high proportion of melanocytic lesions (about 2/3) making overall comparisons of NNB to an average PCP practice inaccurate. Other limitations of our study included a patient population that was only 12.7% Fitzpatrick V-VI.

Conclusions

The DermaSensor device is an easy-to-use, point-of-care, hand-held, AI-developed, algorithm-powered skin cancer adjunctive diagnostic device with high sensitivity and NPV for use in the primary care setting. PCPs can use the device to prioritize skin lesions that merit further evaluation and those that may be monitored.

Footnotes

Acknowledgements

We commend Joseph Massaro, PhD (deceased), for writing the initial study statistical plan and for his contributions to the study design. We acknowledge Irving Bigio, PhD, Eladio Rodriguez-Diaz, PhD, and Preston Hoang, MS, for their contributions developing the device technology. We thank Na Wang, PhD, Margaret Shea, Kanisha Mittal, Lindsey Furton, Kaitlin Hartlage, Cason Hucks, and Matthew Bullard for their contributions to the statistical analyses. We also thank Phillip Leboit, MD, Maxwell Fung, MD, and Jennifer McNiff, MD, for serving as expert dermatopathology panelists for this study. All persons acknowledged above or their institutions were compensated for their contributions to the study. We especially thank all study participants and investigators; without their participation, this study would not have been possible.

Cody Simmons, MS (cofounder and chief executive officer of DermaSensor, Inc), and Kiran Chatha, MD, MPH (DermaSensor, Inc employee), reviewed the manuscript and provided data verification and validation support. Nisha Badders, PhD, ELS, Mayo Clinic, provided editorial suggestions on an earlier draft of the manuscript

Authors’ Note

Presentations

Subsets of the data were presented as posters at (1) the 2nd Annual Innovations in Dermatology Fall Conference; November 3-5, 2022; Las Vegas, Nevada; (2) Maui Derm Hawaii 2023; January 23-27, 2023; Maui, Hawaii; presented as (3) a podium presentation at the 11th Annual American Dermoscopy Meeting; July 13-15, 2023; Stowe, Vermont; (4) as a podium presentation at the American Academy of Dermatology Innovation Academy Meeting; August 10-13, 2023; Tampa, Florida; and (5) the Skin of Color Update conference; October 6-8, 2023; New York, New York. Another subset of the data was presented as a podium presentation at the Society of Teachers of Family Medicine (STFM) Annual Spring Conference; April 29-May 3, 2023; Tampa, Florida.

Portions of this manuscript have been published in abstract form: Cutis. 2022;110(6 Suppl):31; J Clin Aesthet Dermatol. 2023;16(4 Suppl 1):S16; and J Am Acad Dermatol. 2023;89(3 Suppl):AB233.

ORCID iDs

Stephen P. Merry

Ivana T. Croghan

Kimberly A. Dukes

Brian C. McCormick

Gerard T. Considine

Curtis T. Thompson

David J. Leffell

Ethics and Considerations

This study was reviewed and approved by the Western Institutional Review Board-Copernicus Group Review Board (#20182730), the Australian Bellberry Human Research Ethics Committee (#2018-08-620), and the Mayo Clinic Institutional Review Board (#20-011463).

Consent to Participate

Before enrollment, written informed consent was obtained for all participants. Study participants received a small stipend.

Author Contributions

Concept and design: Leffell

Acquisition of data: Merry, Croghan, McCormick, Considine, Thompson

Analysis and interpretation of data: Dukes, Merry, Croghan, Leffell

Drafting of the manuscript: Merry, Croghan

Critical revision of manuscript: Duvall, McCormick, Considine, Thompson, Leffell, Dukes

Statistical analysis: Dukes

Obtained funding: The institutions of Dukes, Merry, Croghan, Duvall, McCormick, Considine, and Thompson received funding for conducting the clinical trial

Administrative, technical, and material support: Thompson

Study supervision: Leffell

No use of artificial intelligence-assisted technologies were used in the production of this manuscript.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by DermaSensor, Inc. The sponsor designed the study and hired a contract research organization to oversee the study conduct and data collection and management. Data analysis was completed by an external biostatistics group at Boston University. The sponsor did not prepare the manuscript but did review it.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Drs Merry and Croghan conducted the study at Mayo Clinic in Rochester, Minnesota, with the support of the sponsor DermaSensor, Inc. Dr Dukes performed statistical analysis under contract from the sponsor at Boston University School of Public Health. Drs McCormick and Considine were site principal investigators at their respective family medicine private practices in Virginia and Australia. Dr Thompson performed dermatopathologic analyses under contract. None have received personal remuneration from the company. Dr Leffell is a member of the DermaSensor Scientific Advisory Board and is compensated by the company.

Abbreviations

AI = artificial intelligence

AUROC = area under the receiver operating characteristic

APP = advanced practice provider

BCC = basal cell carcinoma

DERM-SUCCESS = DERM aSensor S t u dy of Primary C are Physician use of E lastic S cattering S pectroscopy on skin lesions suggestive of skin cancer

ESS = elastic scattering spectroscopy

FDA = US Food and Drug Administration

KC = keratinocyte carcinoma

NNB = number needed to biopsy

NPV = negative predictive value

PCP = primary care physician

PPV = positive predictive value

SCC = squamous cell carcinoma

Trial Registry Information

Trial registration: NCT06690086

Data Availability Statement

All data supporting the study findings are contained in this manuscript. Kimberly A. Dukes, PhD, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Stephen P. Merry, MD, MPH, takes responsibility for the integrity of the work as a whole, from inception to published article.

References

American Cancer Society. Cancer Facts & Figures 2022. American Cancer Society. Accessed October 11, 2022. https://www.cancer.org/content/dam/cancer-org/research/cancer-facts-and-statistics/annual-cancer-facts-and-figures/2022/2022-cancer-facts-and-figures.pdf

Rogers

Weinstock

Feldman

Coldiron

BM.

Incidence Estimate of Nonmelanoma Skin Cancer (Keratinocyte Carcinomas) in the U.S. Population, 2012. JAMA Dermatol. 2015;151(10):1081-1086. (In eng). doi:10.1001/jamadermatol.2015.1187

Muzic

Schmitt

Wright

, et al. Incidence and trends of basal cell carcinoma and cutaneous squamous cell carcinoma: a population-based study in Olmsted County, Minnesota, 2000 to 2010. Mayo Clin Proc. 2017;92(6):890-898. (In eng). doi:10.1016/j.mayocp.2017.02.015

Mansouri

Housewright

CD.

The treatment of actinic keratoses-the rule rather than the exception. JAMA Dermatol. 2017;153(11):1200. (In eng). doi:10.1001/jamadermatol.2017.3395

Greenwood

Merry

Boswell

CL.

Skin biopsy techniques. Prim Care. 2022;49(1):1-22. doi:10.1016/j.pop.2021.10.001

Marchetti

Nanda

, et al. Number needed to biopsy ratio and diagnostic accuracy for melanoma detection. J Am Acad Dermatol. 2020;83(3):780-787. doi:10.1016/j.jaad.2020.04.109

Nelson

Swetter

Saboda

Chen

Curiel-Lewandrowski

Evaluation of the number-needed-to-biopsy metric for the diagnosis of cutaneous melanoma: a systematic review and meta-analysis. JAMA Dermatol. 2019;155(10):1167-1174. (In eng). doi:10.1001/jamadermatol.2019.1514

Privalle

Havighurst

Kim

Bennett

YG.

Number of skin biopsies needed per malignancy: comparing the use of skin biopsies among dermatologists and nondermatologist clinicians. J Am Acad Dermatol. 2020;82(1):110-116. doi:10.1016/j.jaad.2019.08.012

Chen

Bravata

Weil

Olkin

A comparison of dermatologists’ and primary care physicians’ accuracy in diagnosing melanoma: a systematic review. Arch Dermatol. 2001;137(12):1627-134. DOI: 10.1001/archderm.137.12.1627.

10.

Argenziano

Puig

Zalaudek

, et al. Dermoscopy improves accuracy of primary care physicians to triage lesions suggestive of skin cancer. J Clin Oncol. 2006;24(12):1877-1882. doi:10.1200/JCO.2005.05.0864

11.

Shahwan

Kimball

AB.

Should we leave the skin biopsies to the dermatologists?

JAMA Dermatol. 2016;152(4):371-372. doi:10.1001/jamadermatol.2015.5051

12.

Rogers

Marino

Dusza

, et al. A clinical aid for detecting skin cancer: the triage amalgamated dermoscopic algorithm (TADA). J Am Board Fam Med. 2016;29(6):694-701. doi:10.3122/jabfm.2016.06.160079

13.

Dinnes

Deeks

Chuchu

, et al. Dermoscopy, with and without visual inspection, for diagnosing melanoma in adults. Cochrane Database Syst Rev. 2018;12(12):Cd011902. (In eng). doi:10.1002/14651858.CD011902.pub2

14.

Seiverling

Ahrns

Greene

, et al. Teaching benign skin lesions as a strategy to improve the triage amalgamated dermoscopic algorithm (TADA). J Am Board Fam Med. 2019;32(1):96-102. doi:10.3122/jabfm.2019.01.180049

15.

Sawyers

Wigle

Marghoob

Blum

Dermoscopy training effect on diagnostic accuracy of skin lesions in canadian family medicine physicians using the triage amalgamated dermoscopic algorithm. Dermatol Pract Concept. 2020;10(2):e2020035. doi:10.5826/dpc.1002a35

16.

Westerhoff

McCarthy

Menzies

SW.

Increase in the sensitivity for melanoma diagnosis by primary care physicians using skin surface microscopy. Br J Dermatol. 2000;143(5):1016-1020. doi:10.1046/j.1365-2133.2000.03836.x

17.

Menzies

Emery

Staples

, et al. Impact of dermoscopy and short-term sequential digital dermoscopy imaging for the management of pigmented lesions in primary care: a sequential intervention trial. Br J Dermatol. 2009;161(6):1270-1277. doi:10.1111/j.1365-2133.2009.09374.x

18.

Morris

Alfonso

Hernandez

Fernandez

MI.

Examining the factors associated with past and present dermoscopy use among family physicians. Dermatol Pract Concept. 2017;7(4):63-70. doi:10.5826/dpc.0704a13

19.

Fee

McGrady

Rosendahl

Hart

ND.

Dermoscopy use in primary care: a scoping review. Dermatol Pract Concept. 2019;9(2):98-104. (In eng). doi:10.5826/dpc.0902a04

20.

Brown

Najmi

Duke

Grabell

Koshelev

Nelson

KC.

Skin cancer education interventions for primary care providers: a scoping review. J Gen Intern Med. 2022;37(9):2267-2279. doi:10.1007/s11606-022-07501-9

21.

Malvehy

Hauschild

Curiel-Lewandrowski

, et al. Clinical performance of the Nevisense system in cutaneous melanoma detection: an international, multicentre, prospective and blinded clinical trial on efficacy and safety. Br J Dermatol. 2014;171(5):1099-1107. (In eng). doi:10.1111/bjd.13121

22.

Hogarty

Phan

, et al. Artificial intelligence in dermatology-where we are and the way to the future: a review. Am J Clin Dermatol. 2020;21(1):41-47. (In eng). doi:10.1007/s40257-019-00462-6

23.

Beltrami

Brown

Salmon

PJM

Leffell

Grant-Kels

JM.

Artificial intelligence in the detection of skin cancer. J Am Acad Dermatol. 2022;87(6):1336-1342. (In eng). doi:10.1016/j.jaad.2022.08.028

24.

Sangers

Kittler

Blum

, et al. Position statement of the EADV Artificial Intelligence (AI) Task Force on AI-assisted smartphone apps and web-based services for skin disease. J Eur Acad Dermatol Venereol. 2024;38(1):22-30. (In eng). doi:10.1111/jdv.19521

25.

Stafford

Buell

Chiang

, et al. Non-melanoma skin cancer detection in the age of advanced technology: a review. Cancers (Basel). 2023;15(12):3094. (In eng). doi:10.3390/cancers15123094

26.

Valova

Dinh

Gueorguieva

Mobile application for skin cancer classification using deep learning. J Mach Intell Data Sci. 2023;4:15-26. doi:10.11159/jmids.2023.003

27.

Brenes

Silva

Aladab

, et al. The role of smartphone-based applications and artificial intelligence aiding in screening, faster diagnosis and prevention in skin cancer: a systematic mini-review. Principles Pract Clin Res. 2023;9(3):12-17. doi:10.21801/ppcrj.2023.93.4

28.

Sun

Kentley

Mehta

Dusza

Halpern

Rotemberg

Accuracy of commercially available smartphone applications for the detection of melanoma. Br J Dermatol. 2022;186(4):744-746. (In eng). doi:10.1111/bjd.20903

29.

Jahn

Navarini

Cerminara

, et al. Over-detection of melanoma-suspect lesions by a CE-certified smartphone app: performance in comparison to dermatologists, 2D and 3D convolutional neural networks in a prospective data set of 1204 pigmented skin lesions involving patients’ perception. Cancers (Basel). 2022;14(15):3829. (In eng). doi:10.3390/cancers14153829

30.

Smak Gregoor

Sangers

Bakker

, et al. An artificial intelligence based app for skin cancer detection evaluated in a population based setting. NPJ Digit Med. 2023;6(1):90. (In eng). doi:10.1038/s41746-023-00831-w

31.

Rodriguez-Diaz

Manolakos

Christman

, et al. Optical spectroscopy as a method for skin cancer risk assessment. Photochem Photobiol. 2019;95(6):1441-1445. (In eng). doi:10.1111/php.13140

32.

Bigio

Bown

Briggs

, et al. Diagnosis of breast cancer using elastic-scattering spectroscopy: preliminary clinical results. J Biomed Opt. 2000;5(2):221-228. (In eng). doi:10.1117/1.429990

33.

Keshtgar

Chicken

Austwick

, et al. Optical scanning for rapid intraoperative diagnosis of sentinel node metastases in breast cancer. Br J Surg. 2010;97(8):1232-1239. (In eng). doi:10.1002/bjs.7095

34.

Lovat

Johnson

Mackenzie

, et al. Elastic scattering spectroscopy accurately detects high grade dysplasia and cancer in Barrett’s oesophagus. Gut. 2006;55(8):1078-1083. (In eng). doi:10.1136/gut.2005.081497

35.

Mourant

Bigio

Boyer

Conn

Johnson

Shimada

Spectroscopic diagnosis of bladder cancer with elastic light scattering. Lasers Surg Med. 1995;17(4):350-357. (In eng). doi:10.1002/lsm.1900170403

36.

Suh

A’Amar

Rodriguez-Diaz

Lee

Bigio

Rosen

JE.

Elastic light-scattering spectroscopy for discrimination of benign from malignant disease in thyroid nodules. Ann Surg Oncol. 2011;18(5):1300-1305. (In eng). doi:10.1245/s10434-010-1452-y

37.

Stanganelli

Serafini

Bucch

A cancer-registry-assisted evaluation of the accuracy of digital epiluminescence microscopy associated with clinical examination of pigmented skin lesions. Dermatology. 2000;200(1):11-16. doi:10.1159/000018308

38.

Carli

Nardini

Crocetti

De Giorgi

Giannotti

Frequency and characteristics of melanomas missed at a pigmented lesion clinic: a registry-based study. Melanoma Res. 2004;14(5):403-407. doi:10.1097/00008390-200410000-00011

39.

Soyer

Argenziano

Zalaudek

, et al. Three-point checklist of dermoscopy. A new screening method for early detection of melanoma. Dermatology. 2004;208(1):27-31. doi:10.1159/000075042

40.

Dinnes

Deeks

Grainge

, et al. Visual inspection for diagnosing cutaneous melanoma in adults. Cochrane Database Syst Rev. 2018;12(12):CD013194. doi:10.1002/14651858.CD013194

41.

Carney

Reisch

Piepkorn

, et al. Achieving consensus for the histopathologic diagnosis of melanocytic lesions: use of the modified Delphi method. J Cutaneous Pathol 2016;43(10):830-837.

42.

Elmore

Barnhill

Elder

, et al. Pathologists’ diagnosis of invasive melanoma and melanocytic proliferations: observer accuracy and reproducibility study. BMJ 2017;357:j2813. doi:10.1136/bmj.j2813

43.

Saha

Miller

Wang

A comparison of some approximate confidence intervals for a single proportion for clustered binary outcome data. Int J Biostat. 2016;12(2):1-18. (In eng). doi:10.1515/ijb-2015-0024

44.

Simundic

AM.

Measures of diagnostic accuracy: basic definitions. EJIFCC 2009;19(4):203-211. https://www.ncbi.nlm.nih.gov/pubmed/27683318

45.

Hartman

Trepanowski

Chang

, et al. Multicenter prospective blinded melanoma detection study with a handheld elastic scattering spectroscopy device. JAAD Int. 2024;15:24-31. (In eng). doi:10.1016/j.jdin.2023.10.011

46.

Manolakos

Patrick

Geisse

, et al. Use of an elastic-scattering spectroscopy and artificial intelligence device in the assessment of lesions suggestive of skin cancer: a comparative effectiveness study. JAAD Int. 2024;14:52-58. (In eng). doi:10.1016/j.jdin.2023.08.019

47.

Tepedino

Baltazar

Hanna

Bridges

Billot

Zeitouni

NC.

Elastic scattering spectroscopy on patient-selected lesions concerning for skin cancer. J Am Board Fam Med. 2024;37(3):427-435.

Primary Care Physician Use of Elastic Scattering Spectroscopy on Skin Lesions Suggestive of Skin Cancer

Abstract

Objectives:

Patients & Methods:

Results:

Conclusions:

Keywords

Introduction

Methods

Trial Design

Setting

Participants

Device

Study Sequence

Data Analysis

Results

Discussion

Limitations

Conclusions

Footnotes

Acknowledgements

Authors’ Note

ORCID iDs

Ethics and Considerations

Consent to Participate

Author Contributions

Funding

Declaration of Conflicting Interests

Abbreviations

Trial Registry Information

Data Availability Statement

References