BACKGROUND CONTEXT One of the most common exams in a radiologist practice is the lumbar spine. Manual measurements, examination and interpretation are laborious and can vary within and across practices. Artificial intelligence (AI) provides an opportunity to decrease time and has the potential for reducing variability between readers and improving measurement accuracy. PURPOSE This study was initiated to look at imaging data measuring disc height from a clinical population of patients treated for disc degeneration and assess variance between two board-certified radiologists and an investigational AI or auto-segmentation technology (SmartSoft, CoLumbo). STUDY DESIGN/SETTING Patient MRI studies evaluated disc height measurements and calculated disc height change over 12 months. Comparisons between auto-segmentation technology and radiologists looked for differences of not only height but variance in measurement between the auto-segmentation technology and radiologists. PATIENT SAMPLE An initial MRI was specified up to 6 months prior to treatment and a follow-up MRI within 6 months after a 12-month endpoint. Treatments were agnostic to the intradiscal intervention. MRI disc heights were assessed pretreatment and post-treatment at the treated levels for 107 patients. OUTCOME MEASURES A change in disc height was calculated as the difference of the pretreatment height minus the post-treatment height. The average height and the standard deviation were determined for a comparison between the radiologists and auto-segmentation technology. Methods Two board-certified radiologists, blinded to treatment and levels treated, measured all lumbar disc levels at a mid-sagittal central location. A comparison was made between the two radiologists and an auto segmentation software (SmartSoft, Columbo) that read, segmented, and measured the MRI images. The software measured disc height as an average value of three parasagittal measurements for the anterior margin, midcentral sagittal, and the posterior margin. The midcentral sagittal slice was most comparable to the defined criteria used by the radiologists and used for analysis. Statistical analysis was performed using Minitab software (version 19.2020.1). Results Interobserver measurement between the radiologists reflected the complexity and irregularity of the vertebral plate. Radiologist 2 measurements were consistently lower for both pretreatment and post-treatment (mean differences of 0.51 to 0.85 mm). However, the variances of the height measurements for both radiologists across all treatment groups were not statistically different (p = .630; Bartlett's test). The SmartSoft mid-central sagittal measurements, paired to criteria used for the radiologist readings, were statistically different compared to either of the radiologists for treatment groups at pretreatment and at post-treatment (p < .001 for all comparisons at every level measured; one-way ANOVA). When the disc height changes were pooled, a statistically different variance of SmartSoft measurements was ±0.93 mm compared to ±1.21 mm for all radiologist measurements (p <.001; F test). A comparison of SmartSoft vs Radiologist 1 also had a statistically different variance (±0.93mm vs ±1.40mm; p <.001; F test). However, a comparison of SmartSoft vs Radiologist 2 did not have a statistically different variance (±0.93 mm vs ± 1.00 mm; p = .388; F test). Conclusions Disc height measurements and the calculated disc height change over 12 months compared auto-segmentation technology with two board-certified radiologists. Interobserver variability was noted between the two radiologist's measurements. The pooled variance disc height change for the SmartSoft Technology was less than the pooled variance of the radiologists. While radiologist 1 was not significantly different in variance vs the SmartSoft Technology, the other radiologist was significantly different. SmartSoft Technology could provide a more consistent measurement as the variance in a clinical study assessment where MRI and disc height are measured might be critical. Separately, the acquisition and measurement automation yields a less variable assessment with fractional time commitment. FDA DEVICE/DRUG STATUS This abstract does not discuss or include any applicable devices or drugs. One of the most common exams in a radiologist practice is the lumbar spine. Manual measurements, examination and interpretation are laborious and can vary within and across practices. Artificial intelligence (AI) provides an opportunity to decrease time and has the potential for reducing variability between readers and improving measurement accuracy. This study was initiated to look at imaging data measuring disc height from a clinical population of patients treated for disc degeneration and assess variance between two board-certified radiologists and an investigational AI or auto-segmentation technology (SmartSoft, CoLumbo). Patient MRI studies evaluated disc height measurements and calculated disc height change over 12 months. Comparisons between auto-segmentation technology and radiologists looked for differences of not only height but variance in measurement between the auto-segmentation technology and radiologists. An initial MRI was specified up to 6 months prior to treatment and a follow-up MRI within 6 months after a 12-month endpoint. Treatments were agnostic to the intradiscal intervention. MRI disc heights were assessed pretreatment and post-treatment at the treated levels for 107 patients. A change in disc height was calculated as the difference of the pretreatment height minus the post-treatment height. The average height and the standard deviation were determined for a comparison between the radiologists and auto-segmentation technology. Two board-certified radiologists, blinded to treatment and levels treated, measured all lumbar disc levels at a mid-sagittal central location. A comparison was made between the two radiologists and an auto segmentation software (SmartSoft, Columbo) that read, segmented, and measured the MRI images. The software measured disc height as an average value of three parasagittal measurements for the anterior margin, midcentral sagittal, and the posterior margin. The midcentral sagittal slice was most comparable to the defined criteria used by the radiologists and used for analysis. Statistical analysis was performed using Minitab software (version 19.2020.1). Interobserver measurement between the radiologists reflected the complexity and irregularity of the vertebral plate. Radiologist 2 measurements were consistently lower for both pretreatment and post-treatment (mean differences of 0.51 to 0.85 mm). However, the variances of the height measurements for both radiologists across all treatment groups were not statistically different (p = .630; Bartlett's test). The SmartSoft mid-central sagittal measurements, paired to criteria used for the radiologist readings, were statistically different compared to either of the radiologists for treatment groups at pretreatment and at post-treatment (p < .001 for all comparisons at every level measured; one-way ANOVA). When the disc height changes were pooled, a statistically different variance of SmartSoft measurements was ±0.93 mm compared to ±1.21 mm for all radiologist measurements (p <.001; F test). A comparison of SmartSoft vs Radiologist 1 also had a statistically different variance (±0.93mm vs ±1.40mm; p <.001; F test). However, a comparison of SmartSoft vs Radiologist 2 did not have a statistically different variance (±0.93 mm vs ± 1.00 mm; p = .388; F test). Disc height measurements and the calculated disc height change over 12 months compared auto-segmentation technology with two board-certified radiologists. Interobserver variability was noted between the two radiologist's measurements. The pooled variance disc height change for the SmartSoft Technology was less than the pooled variance of the radiologists. While radiologist 1 was not significantly different in variance vs the SmartSoft Technology, the other radiologist was significantly different. SmartSoft Technology could provide a more consistent measurement as the variance in a clinical study assessment where MRI and disc height are measured might be critical. Separately, the acquisition and measurement automation yields a less variable assessment with fractional time commitment.