The <scp>AI</scp> Innerve Algorithm: Automatic Reproducible Reading for Intraepidermal Nerve Fiber Density Analysis, a New Tool for Small‐Fiber Neuropathy Diagnosis

OBJECTIVES: To measure intra-observer and interobserver correlations and reading times for visual and artificial intelligence algorithm-assisted reading (innerve) of intraepidermal small nerve fiber density (IENFD) in skin biopsies. METHODS: We retrospectively selected 40 skin biopsy slides with PGP9.5/col4 double-immunofluorescence labeling. The slides were scanned and read by two experienced observers (O1 and O2), on two occasions (visual reading 1 -VR1- and 2 -VR2-). They were then read after predetection with the Innerve algorithm (assisted reading -AR-). IENFD and reading time were evaluated to assess reliability and time saved. RESULTS: Intra-observer reliability was evaluated by calculating intraclass correlation coefficient (ICC). The ICC obtained were 0.95 (O1)/0.99 (O2) for the two VR and 0.96 (O1)/0.99 (O2) for VR1/AR. Inter-rater ICC between O1 and O2 was 0.88, 0.87, and 0.93 for VR1, VR2, and AR, respectively. AR reading time was 1.96 times faster for O1 and 1.03 times slower for O2. Subgroup analyses revealed an interobserver ICC significantly > 0.7 (p < 0.05), with ICC > 0.9 regardless of sex and skin pigmentation. DISCUSSION: Innerve is an innovative and reliable AI tool for IENFD determinations on skin biopsies, making such analyses-which were considered time-consuming and difficult-easier for small-fiber neuropathy diagnosis.