Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1679
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-17T10:41:40.202088Z
status: published
lens: trace
sector: health
headline: Accuracy of general-purpose large language models (LLMS) in interpreting adolescent idiopathic scoliosis radiographs: a comparative study with experienced spine surgeons
dek: Purpose: This study evaluates the accuracy of general-purpose, vision-capable large language models (LLMs) in interpreting adolescent idiopathic scoliosis (AIS) radiographs and compares their outputs with an expert surgeon consensus reference standard.
Methods: This retrospective comparative diagnostic accuracy study analyzed 128 anonymized AIS radiographic case sets from 128 different patients, including standing coronal and sagittal radiographs and lateral bending films. ChatGPT GPT-5.2 Instant, Gemini 3 Flash…
gain_title: General-purpose vision LLMs estimated scoliosis Cobb angles within 5 degrees of spine surgeon consensus in most cases, with ChatGPT agreeing in 94.5% of the 128 cases.
problem_title: The same models showed substantially lower exact agreement on Lenke classification, modifiers, and fusion-related landmarks, so they are not reliable for comprehensive radiograph interpretation or surgical planning.
trace_subject: General-purpose LLM interpretation of adolescent idiopathic scoliosis radiographs
gain_reading: General-purpose vision LLMs estimated scoliosis Cobb angles within 5 degrees of spine surgeon consensus in most cases, with ChatGPT agreeing in 94.5% of the 128 cases.
gain_evidence: ChatGPT achieved 94.5% agreement within ± 5° of surgeon consensus, with a mean absolute CA error of 2.4°
problem_reading: The same models showed substantially lower exact agreement on Lenke classification, modifiers, and fusion-related landmarks, so they are not reliable for comprehensive radiograph interpretation or surgical planning.
problem_evidence: Exact agreement was substantially lower for Lenke classification, lumbar and sagittal modifiers, neutral vertebra, and stable vertebra identification | The findings do not support independent use of these systems for scoliosis measurement, education, or surgical planning
quick_read: Researchers gave 128 anonymized radiograph sets from adolescents with idiopathic scoliosis to three general-purpose AI chatbots, ChatGPT, Gemini and Grok, and compared their outputs with the consensus of three experienced spine surgeons. The images had no measurement lines or labels, and the models were prompted using the Lenke classification system.

The results suggest these tools can approximate curve size but cannot yet be trusted for fuller radiographic reading that guides surgery. The authors frame them only as supervised documentation or teaching aids. The study tested specific model versions in January 2026 and did not formally rank the models against one another.
limitation: Only specific versions of three general-purpose models were tested in January 2026, and no formal inferential ranking between the models was performed.
tag: Dual reading
key_points: ChatGPT reached 94.5% agreement within 5 degrees of surgeon consensus on Cobb angle, versus 92.2% for Gemini and 90.6% for Grok. | Mean absolute Cobb angle errors were 2.4 degrees for ChatGPT, 2.7 for Gemini and 3.1 for Grok. | Exact agreement was substantially lower for Lenke classification, lumbar and sagittal modifiers, neutral vertebra and stable vertebra identification. | Agreement among the three surgeons was high for Cobb angle and categorical parameters.
rundown: The retrospective diagnostic accuracy study used standing coronal and sagittal radiographs plus lateral bending films from 128 different patients. Models were accessed through official web interfaces using one standardized prompt, with de-identified raw images lacking Cobb values, vertebral labels or Surgimap overlays.

Three surgeons with more than 10 years of deformity surgery experience independently assessed every case using Surgimap, and their consensus served as the reference standard. Cobb angle accuracy was judged by mean absolute error and agreement within 5 degrees, while categorical outputs were judged by exact agreement with 95% confidence intervals.

The authors conclude that performance was weaker for fusion-level-related landmarks than for angle estimation and that the tools should not be used for independent measurement, education or surgical planning.
sources:
- peer_reviewed | European Spine Journal | https://doi.org/10.1007/s00586-026-10443-z | 2026-10-16
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

8eaab90a26692bb58c5d0fdfd8b4be228f8d38cf4d3298c82b2946b03e2521e3

previous

0000000000000000000000000000000000000000000000000000000000000000