74. Novel artificial intelligence algorithm can accurately and independently measure spinopelvic parameters

Abstract

<h3>BACKGROUND CONTEXT</h3> Preoperative and postoperative sagittal plane assessment is crucial in both spinal deformity and degenerative pathologies. Sagittal malalignment is a well-established cause of poor patient reported outcomes. There is a growing need for an automated analysis tool that measures pelvic parameters with speed, precision and reproducibility without relying on user identified landmarks. A new AI algorithm has been developed to measure important radiographic parameters independently. <h3>PURPOSE</h3> The analysis of sagittal alignment by measuring spinopelvic parameters has been widely adopted among spine surgeons globally. The currently available spine measurement software programs require users to identify several landmarks prior to calculating parameters, making them time consuming and more reliant upon user experience. This study evaluates and demonstrates that an algorithm based on artificial intelligence (AI) can independently determine spinopelvic parameters. This allows for precise radiographic measurement without time-intensive human input. We hypothesize that the novel, fully automatic method will have a high agreement with human measurements for lumbar lordosis (LL), pelvic incidence (PI), pelvic tilt (PT) and sacral slope (SS). <h3>STUDY DESIGN/SETTING</h3> Evaluation of the inter-rater reliability and mean error between radiographic measurements of the AI algorithm and expert human raters. Radiographs were retrospectively reviewed from a single site, multisurgeon center. <h3>PATIENT SAMPLE</h3> In this study, preoperative and postoperative X-rays were evaluated in 100 patients undergoing a lumbar fusion procedure. <h3>METHODS</h3> From a total of 200 lateral lumbar radiographs (preoperative and postoperative images from 100 patients undergoing fusion) 5 independent observers (4 spinal surgeons, 1 senior researcher) digitally measured LL, PI, PT and SS. Their parameters were compared with AI algorithm generated parameters. Mean error (95% confidence interval, standard deviation) and inter-rater reliability were assessed using 2-way mixed, single-measure intraclass correlation (ICC). ICC values larger than 0.75 were considered excellent. <h3>RESULTS</h3> The novel algorithm's spinopelvic parameter ICC values were excellent in 98% of preoperative and in 95% of postoperative radiographs (PreOp range: 0.85–0.92, PostOp range: 0.81–0.87). Exemplarily, mean errors are smallest for the PI (PreOp: -0.5° [95%-CI: -1.5°–0.6°]; PostOp: 0.0° [-1.1°–1.2°]) and largest for LL (1.3° [0.3°–2.4°]; 3.8° [2.5°–5.0°]). <h3>CONCLUSIONS</h3> Novel AI algorithm automated spinopelvic parameter measurements from spine radiographs have a high degree of accuracy comparable to digital measurements by experts. This algorithm can improve physician workflow efficiency and reduce inter-rater and intra-rater measurement errors <h3>FDA DEVICE/DRUG STATUS</h3> This abstract does not discuss or include any applicable devices or drugs.