당뇨병 발 평가를 위한 열화상 카메라 검증 (HIK-DM-2026)

Each participant is assessed in a single session in a thermally stable room at 20 to 25 degrees Celsius, with relative humidity recorded as a covariate. After 15 minutes of acclimatization in a semi-Fowler position with the backrest at 30 to 45 degrees, both cameras acquire plantar thermograms from a static tripod at a fixed distance of 75 cm, with the tripod head at 90 degrees to the plantar plane and an emissivity setting of 0.98. The two radiometric acquisitions are consecutive and always less than 10 seconds apart. The order of the cameras is randomized 1:1 at participant level (sequence A: M60 then Mini2 V2; sequence B: Mini2 V2 then M60), from a list generated in advance by a team member not involved in image acquisition and kept in a password-protected file until recruitment is complete. No specific background is used.

Thermograms are exported to HIKMICRO Analyzer version 2.0.1. Six anatomical regions of interest, defined a priori and identical for both cameras, are delimited on each foot: hallux, first, third and fifth metatarsal heads, midfoot and heel. There is no post-hoc region selection and no indexing to either device. Contralateral thermal asymmetry (Delta T) is computed per region as the absolute difference in mean temperature between homologous regions of the right and left foot.

Images are renamed with an alphanumeric code that removes any explicit reference to the source device before analysis, and regions are delimited by an independent evaluator blinded to the neurological status of the participant. Technical blinding is acknowledged as partial, because the difference in spatial resolution between the two cameras may allow a trained evaluator to infer the source device.

The sample size follows the method of Walter, Eliasziw and Donner for the intraclass correlation coefficient, with a null hypothesis of ICC0 = 0.65 against an alternative of ICC1 = 0.80, alpha = 0.05, nominal power 0.80 and two observations per participant, which yields a minimum of 77 participants. The study enrolls 130 participants, which raises the actual power for this contrast to 95.7 percent and absorbs expected losses.