Zero-Cycle: First Continuous, Unsupervised, At-Home Monitoring of Spinal Motion as a Per-Cycle Substrate for Machine Learning

Continuous, objective records of spinal motion during ordinary life have not been available; assessment relies on intermittent questionnaires and brief supervised motion capture. A dual inertial-sensor garment applies the Zero-Cycle method, which organises the differential thorax–pelvis motion into movement cycles and emits one fixed-dimensional feature vector per cycle, using recurrent movement configurations as body-intrinsic anchors. In the implementation used for this study, the differential orientation is projected onto a horizontal plane at a nominal 50 cm scale, centred on a data-derived hub, and segmented at recurrent radial minima; each cycle contributes a start landmark, an apex landmark and their difference, in centimetres of projected radius. No magnetometer is used, and no calibration posture is required at any point. In a four-patient feasibility pilot conducted unsupervised at home around two foraminotomies and two transforaminal epidural steroid injections, 88,027 movement cycles were analysed over 13 to 16 recorded days per patient, across 87 recording-defined sessions during patient-managed home wear. Because home life is unscripted, the principal daily display is Best-200, the lowest-excursion window of 200 consecutive retained cycles observed that day, read as a candidate estimate of the day’s best-controlled state following the best-of-repeated-efforts convention of standardised clinical measurement; whole-day and alternative summaries are reported alongside. The two surgical patients showed a reduction that persisted under every summary rule (cycle-level Cliff’s delta +0.93 and +0.62); the comparison injection case showed no comparable reduction despite substantial improvement in patient-reported outcomes; the index injection case was summary-dependent. Excursion and return-related quantities are motivated by Dubousset’s cone-of-economy and Panjabi’s neutral-zone concepts, without claiming equivalence to either. The contribution is a substrate for machine learning rather than a more accurate sensor; no model is trained and no treatment effect is established.