FDA 510(k) Search - By Innolitics (SaMD and AI Experts)

AI/MLSaMDIVD (In Vitro Diagnostic)TherapeuticDiagnosticis PCCP AuthorizedThirdpartyExpeditedreview

Intended Use

The DermaSensor device is indicated for use to evaluate skin lesions suggestive of melanoma, basal cell carcinoma, and/or squamous cell carcinoma in patients aged 40 and above to assist in the decision regarding referral of the patient to a dermatologist. The DermaSensor device should be used in conjunction with the totality of clinically relevant information from the clinical assessment, including visual analysis of the lesion, by physicians who are not dermatologists. The device should be used on lesions already assessed as suspicious for skin cancer and not as a screening tool. The device should not be used as the sole diagnostic criterion nor to confirm clinical diagnosis of skin cancer.

Device Description

The DermaSensor device (hereinafter referred to as 'DermaSensor', or the 'DermaSensor device'; Figure 1) utilizes optical spectroscopy and an artificial intelligence/machine learning (AI/ML) based software algorithm to analyze an intact skin lesion to which the device is non-invasively applied.

The device is a combination of a handheld unit and a base unit. The handheld unit contains a xenon arc lamp and a fiber-optic probe tip which together transmit broadband white light to a lesion surface. Samples of the backscattered light from the tissue are collected by an adjacent detection optical fiber, also within the probe tip, and are conveyed to a microspectrometer, vielding Elastic Scattering Spectroscopy (ESS) spectral recordings. The handheld unit is operated using a touchscreen interface with step-by-step guidance. The small fiber-optic tip is the only component that contacts the patient. The handheld unit remains in the base when not actively being applied to a lesion and its battery is recharged by the base's wireless charging mechanism. The base unit also contains calibration material that is accessible to the handheld unit.

In the DermaSensor device, analysis of the optical recordings of backscattered light over the range of wavelengths is carried out using a proprietary ML-derived classifier algorithm. The spectrum of scattered intensity vs. wavelength is a pattern, which is analyzed by a proprietary classifier algorithm in the device's built-in microprocessor to assess for the potential presence of melanoma. squamous cell carcinoma, or basal cell carcinoma. An internal microprocessor and classifier algorithm analyze the recording and provide results to the user at the point of care, Results are provided as "Monitor," for a negative result, or "Investigate Further" for a positive result. For positive output ("Investigate Further") the DermaSensor additionally displays a Similarity Score of 1-10, with higher scores representing greater similarity to signals seen in malignant lesions.

AI/ML Overview

Here's a breakdown of the acceptance criteria and study proving device performance for the DermaSensor, based on the provided text:

Acceptance Criteria and Reported Device Performance

Note: The document outlines the special controls (acceptance criteria) that the device must meet. The "Reported Device Performance" column reflects the results presented in the CLINICAL PERFORMANCE TESTING section, specifically from the DERM-SUCCESS pivotal study and the reader studies.

Acceptance Criteria Category (Special Control)	Specific Acceptance Criteria	Reported Device Performance
1. Clinical Performance Validation Testing (Premarket & Post-market)	(i) Data must demonstrate superior accuracy of device-aided users' diagnostic characterization of the indicated lesions compared to the accuracy of unaided users.
(iii) Standalone device performance testing must demonstrate the accuracy of the device output relative to ground truth, including:
(A) Testing must demonstrate at least 90% sensitivity of the device output for lesions with high metastatic potential, or an alternative clinical consideration must be provided to justify lower sensitivity. Clinical justification must be provided for the reported specificity.	Device-Aided User Performance (MRMC Studies):

DERM-SUCCESS initial reader study (all skin cancers): Device-aided PCPs had a sensitivity of 91.4% (vs. unaided 82.0%; p=0.0027). Sensitivity + specificity > 1. AUROC increased 5.4% (from 0.708 to 0.762).
DERM-SUCCESS pigmented lesion reader study (pigmented lesions): Aided PCP AUROC > unaided PCP AUROC by 1.5% (p unaided AUROC by 4.1% (p 1 (95.5% + 20.7% > 1).
Met secondary endpoint of overall sensitivity > 90% (p90%) for positive binary decisions. Deemed acceptable within clinical performance context. |
| 4. Electrical Safety, Mechanical Safety, Thermal Safety, EMC | Performance testing must demonstrate the electrical safety, mechanical safety, thermal safety, and electromagnetic compatibility of any electrical components of the device. | Passed all relevant testing according to ETSI EN 301 489-17, IEC 60601-1-2, IEC 60601-1, IEC 62133-2, IEC 62471, IEC 60601-1-6, and IEC 60601-2-57. |
| 5. Reprocessing Instructions Validation | Performance testing must validate reprocessing instructions for reusable components of the device. | Cleaning and disinfection validations conducted per FDA guidance, AAMI TIR 12:2010, and AAMI TIR30:2011 (R2016). All testing and results adequate. |
| 6. Biocompatibility | The patient-contacting components of the device must be demonstrated to be biocompatible. | Passed cytotoxicity (No evidence of cell lysis or toxicity), sensitization (met requirements at 24, 48, 72 hrs), and irritation (No erythema or edema observed at 24, 48, 72 hrs) tests according to ISO 10993-1, -5, -10, -12 and FDA guidance. |
| 7. Software Verification, Validation, and Hazard Analysis | Must be performed. | Software developed and tested according to FDA guidance (2005, 2014, 2023), IEC 62304, and ISO 14971. Documentation and testing (including cybersecurity) demonstrate proper operation, address potential hazards (malfunction, errors, hardware failures, unauthorized access) with satisfactory results. Adequate for reasonable assurance of specified operation and protection from cyber vulnerability. |
| 8. Human Factors Assessment | Must demonstrate that the device can be safely and correctly used by intended users. | Summative human factors validation testing conducted with two user groups (15 PCPs, 15 mid-level practitioners). All 30 participants identified appropriate indications/contraindications and acknowledged the device is not for sole diagnosis. No unmitigated use difficulties, close calls, or use errors. Results demonstrated safe and effective operation by intended users. |
| 9. Labeling | (i) Summary of standalone and clinical performance, including sensitivity, specificity, and CIs for relevant subgroups.
(ii) Description of patient population used in algorithm development/training.
(iii) Device limitations or subpopulations where performance may differ/not validated.
(iv) Interpretation information, including risks of misinterpretation.
(v) Warnings for energy-emitting components.
(vi) Statement: "not intended for standalone diagnostic."
(vii) Maintenance/reprocessing instructions. | Labeling confirmed to include: performance measures (sensitivity, specificity, CI), performance for subgroups, patient population for algorithm training/development, limitations (e.g., increased risk patients, lesion types/locations, Fitzpatrick IV-VI limitations), specific warnings, statement that it is not intended for standalone diagnostic, and maintenance/reprocessing instructions. Post-market surveillance will update labeling. |

Study Proving Device Meets Acceptance Criteria

The primary study proving the device meets the acceptance criteria is the DERM-SUCCESS pivotal clinical study, supplemented by three Multi-Reader, Multi-Case (MRMC) comparative effectiveness studies.

1. A table of acceptance criteria and the reported device performance

(See table above)

2. Sample size used for the test set and the data provenance

Test Set (DERM-SUCCESS pivotal study):

Sample Size: 1,005 participants with 1,579 lesions (from an initial enrolment of 1,021 participants with 1,598 lesions).
Data Provenance:
- Country of Origin: International, multicenter study conducted at 22 study sites.
  - 18 locations in the United States (including Arizona, California, Florida, Kansas, Minnesota, Pennsylvania, Rhode Island, Tennessee, Texas, Utah, Virginia),
  - 4 locations in Australia.
- Retrospective or Prospective: Prospective, blinded clinical study.

Test Set (MRMC Reader Studies):

DERM-SUCCESS initial reader study: 25 malignant and 25 benign lesions (total 50 lesions).
DERM-SUCCESS pigmented lesion reader study: 69 malignant lesions (36% melanoma, 26% SCC, 36% BCC) and 67 benign lesions (total 136 lesions).
DERM-ASSESS III reader study: 50 malignant lesions (68% melanoma, 16% SCC, 16% BCC) and 50 benign lesions (total 100 lesions).

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts

DERM-SUCCESS pivotal study (Standalone Performance):

Number of Experts: 2 to 5 central study dermatopathologists.
Qualifications of Experts: Dermatopathologists. (Specific experience level like "10 years of experience" is not mentioned, but the term implies specialized expertise in skin pathology).

MRMC Reader Studies (Reader Performance Evaluation):

Number of Experts: Not directly stated for ground truth, but the lesions were derived from the DERM-SUCCESS pivotal study, implying the same ground truth establishment method.
For Reader Participants:
- DERM-SUCCESS initial reader study: 108 readers trained in internal medicine or family practice (PCPs).
- DERM-SUCCESS pigmented lesion reader study: 77 PCP readers.
- DERM-ASSESS III reader study: 118 PCP readers.

4. Adjudication method for the test set

DERM-SUCCESS pivotal study:

Pathology findings were validated by two to five central study dermatopathologists depending on the histological severity and discordance for the diagnoses. This implies a consensus or majority vote approach for adjudication, with more experts involved as complexity or disagreement increased.

5. If a multi reader multi case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance

Yes, three MRMC comparative effectiveness studies were done:

DERM-SUCCESS initial reader study (all skin cancers):
- Effect Size: Device-aided PCPs had a higher sensitivity of 91.4% compared with unaided PCP sensitivity of 82.0% (an increase of 9.4% points).
- The AUROC (overall performance) increased 5.4% (from 0.708 to 0.762).
DERM-SUCCESS pigmented lesion reader study:
- Effect Size (AUROC): Aided PCP AUROC was greater than unaided PCP AUROC by 1.5%.
- Effect Size (Sensitivity, all skin cancers): Aided PCP sensitivity increased from 80.5% to 86.3% (an increase of 5.8% points).
- Effect Size (Sensitivity, melanoma): Aided PCP sensitivity for melanoma increased from 68.8% to 75.4% (an increase of 6.6% points).
DERM-ASSESS III reader study (lesions suggestive of melanoma):
- Effect Size (AUROC): Aided AUROC was greater than unaided AUROC by 4.1%.
- Effect Size (Sensitivity): Aided sensitivity increased from 73.7% to 81.8% (an increase of 8.1% points).
- Effect Size (Sensitivity, melanoma): Aided sensitivity for melanoma increased from 70.2% to 79.1% (an increase of 8.9% points).

6. If a standalone (i.e. algorithm only without human-in-the loop performance) was done

Yes, a standalone performance evaluation was done in the DERM-SUCCESS pivotal clinical study.

Overall Sensitivity (High-Risk Lesions): 95.5% (91.7%-97.6%) for identification of melanoma, SCC, or BCC.
Overall Specificity (Benign Lesions): 20.7% (18.5%-23.1%).
AUROC: 0.7896.

7. The type of ground truth used

The ground truth used was pathology findings (histological diagnosis). These findings were validated by multiple central study dermatopathologists.

8. The sample size for the training set

The training dataset initially contained 950 lesions contributing approximately 4,200 spectra. This study dataset was later used entirely for algorithm training during the initial development phases.

A separate table provides detailed demographics for a "Training Dataset" and a "Tuning Dataset":

Training Dataset (Table 2): 1067 lesions (sum of age groups, excluding unknowns)
Tuning Dataset (Table 2): 438 lesions (sum of age groups)
This suggests a larger training and tuning dataset was used for the final algorithm, totaling 1505 lesions.

9. How the ground truth for the training set was established

The text describes initial algorithm development using a "blinded, multicenter clinical study through an initial ML retrospective algorithm." The ground truth for both the training and testing datasets in this earlier study was established by evaluating "the ability of ESS to differentiate malignant from benign melanocytic lesions." While not explicitly stated for this training set, it is highly probable that the ground truth for the training set was established through histological diagnosis (biopsy results), consistent with how the test set ground truth was established, given the nature of skin cancer detection. The subsequent pivotal study (DERM-SUCCESS), which used biopsy as ground truth, built upon this prior development.