ABSTRACT Non-melanoma keratinocyte carcinomas, namely basal cell (BCC) and squamous cell (SCC) carcinomas, represent the most common form of cancer and their incidence is increasing. When diagnosed at the earliest stage, such as superficial forms of BCC or SCC-in-situ confined to the epidermis, they are usually treatable by non-surgical means. Curative therapy for more advanced tumors, such as micronodular or infiltrative BCC subtypes or invasive SCC with extensive dermal involvement, requires surgery that can be disfiguring and costly. Some SCC can resemble actinic keratosis (AK, precancer) or inflamed seborrheic keratosis (SK). Visual detection of these skin cancers and benign keratoses can be facilitated with the aid of dermoscopy, but determining whether a lesion is more deeply invasive or distinguishing between SCC-in-situ, AK and inflamed SK is not always clinically apparent. There is no currently available low cost, non-invasive technique to overcome these diagnostic hurdles to inform whether to biopsy or which biopsy technique (shave for superficial lesions vs. punch for invasive lesions) is most appropriate. Electrical impedance dermography (EID) can measure electrical properties of the skin that are altered in skin cancers, however, whether EID technology can distinguish superficial from invasive subtypes, or SCC-in-situ from actinic and inflamed keratoses, is unknown. We have built and tested a novel machine-learning (ML) augmented EID device, which can quickly perform non-invasive measurements on the skin and requires minimal training for use. We have published studies on cohorts of 17 and 35 subjects with BCC and SCC, respectively, and demonstrated high reproducibility. We have also recently completed a pilot blinded study on cohorts of 35 subjects with BCC, invasive SCC, SCC-in-situ, and SK, and demonstrated superior diagnostic accuracy in distinguishing SCC subtypes from inflamed SK than commercially available devices (Nevisense). Our overarching hypothesis is that EID can distinguish BCC subtypes and between SCC-in-situ, invasive SCC, and benign keratoses. In Aim 1, we will collect EID data in 80 subjects with lesions clinically suspicious for BCC. Then, we will use the EID data to train our pilot ML algorithm, including morphologic and histologic characteristics, to distinguish BCC subtypes. Finally, we will test its performance in a blinded study with a separate test cohort in 40 subjects. In Aim 2, we will expand our SCC/SK cohort to further improve the diagnostic performance of our pilot ML algorithm at distinguishing SCC-in- situ, invasive SCC, AK, and inflamed SK in 80 subjects with lesions clinically suspicious for SCC. Analyses will include standard reproducibility metrics (first vs. second measurement), two-group comparisons, and correlations to histology. Successful application of this EID technology will contribute to overall clinical assessment by increasing diagnostic confidence and guiding decisions to biopsy (and biopsy technique). The data collected will set the stage for validating our findings in a multicenter cohort to evaluate the utility of this approach for non-invasive diagnosis and management of suspicious skin lesions.

PROJECT NARRATIVE Here, we will assess a novel non-invasive electrical impedance dermography screening device with potential to serve as an early diagnostic biomarker in patients with basal cell and squamous cell carcinoma. Data acquired at the completion of this work will support future evaluation to assist dermatologists and primary care physicians to inform biopsy-decision making in patients with lesions suspicious for non-melanoma cancer.