Unlike general image processing software, spinal measurement software falls into auxiliary diagnostic medical devices whose core value lies in delivering accurate and reliable anatomical parameters (including Cobb angle, vertebral rotation, scoliosis severity and so forth) for clinicians. Amid NMPA’s growing evidence-based review tendency, registration of spinal measurement software no longer consists of simple function listing; instead, applicants are required to establish a full evidence chain spanning from algorithm principles to clinical application.
Different from general image processing software, spinal measurement software is categorized as auxiliary diagnostic medical device. Its core value lies in providing clinicians with accurate and reliable anatomical parameters, such as Cobb angle, vertebral rotation degree, and scoliosis severity. Against the backdrop of the National Medical Products Administration’s increasingly emphasized evidence-based review principle, the registration application of spinal measurement software is no longer a simple presentation of product functions. Instead, enterprises are required to build a complete evidence chain covering algorithm principles to clinical application.
It is critical for project teams preparing for Class II/Class III medical device registration and clinical evaluation to clarify the logical thinking of technical reviewers. Only by deeply integrating clinical measurement requirements with rigorous software engineering specifications can enterprises substantially reduce the risk of supplementary data requests and ensure smooth project approval.
During technical review, experts focus far beyond the numerical deviation between software-measured angles and the golden standard; core attention lies in the sustainability, stability and safety of measurement outcomes under real clinical use. For spinal measurement software, the following three dimensions serve as top review priorities:
Review experts disapprove of unexplainable black-box algorithms. If the software automatically recognizes vertebral borders and calculates Cobb angles, its identification logic shall be explicitly specified, whether derived from edge detection algorithms (e.g. Canny), deep learning frameworks (e.g. U-Net) or conventional template matching.
Key Point: Algorithm schematic diagram is mandatory to elaborate the full workflow from image input, feature landmark extraction to final angle calculation. For AI-involved products, evidence shall be provided to verify the model can truly identify vertebra anatomy rather than merely fit pixel distribution.
What is the commonly accepted golden standard for radiological measurement in Western medicine? Manual measurements by senior orthopedic specialists or high-precision 3D CT reconstruction?
Key Point: General statement of "comparison against specialist readings" is unacceptable. The reference standard shall be precisely defined, such as "average value from independent measurements by three orthopedic physicians with associate chief physician title or above" or "actual intraoperative navigation readings", alongside detailed formulation rationale to guarantee fair comparison.
High precision under laboratory settings cannot guarantee clinical applicability. Spinal X-ray images frequently suffer from improper shooting angles, irregular patient positioning, image noise and shielding caused by metallic implants.
Key Point: Reviewers require test results under adverse clinical conditions. The verification dataset shall consist of real-world data from multiple equipment manufacturers (GE, Siemens, etc.), various projection modes (AP, lateral, oblique), and diverse pathological conditions including severe scoliosis and postoperative follow-up cases.
To pass technical review, project teams shall follow the below document preparation practices matching reviewers' reading habits:
The intended use of the software must be clarified at the early application stage. Confirm whether the product serves merely as a measurement tool or includes auxiliary diagnosis functions. Products functioning only as measuring rulers are generally classified as Class II devices; those automatically outputting scoliosis diagnosis suggestions or surgical recommendations based on measured values fall into high-risk CDSS (Clinical Decision Support Software) under Class III administration. Confirm whether the product is standalone software or an embedded component of imaging equipment. Standalone software submissions require cybersecurity and version management documentation; accessory software bundled with X-ray devices shall be filed together with corresponding hardware as embedded software components.
Reviewers require verification to prove stable consistent performance rather than merely executable algorithm operation.
Prioritize data compliance: All data for algorithm training and testing shall be legally sourced; labeling shall involve professional physicians with complete labeling specifications and ethical approval documents retained.
Quantify performance metrics: Avoid vague descriptions such as "accurate measurement". Specify statistical indicators including MAE (Mean Absolute Error), RMSE (Root Mean Square Error), ICC (Intraclass Correlation Coefficient), etc. For instance, Cobb angle measurement error shall be confined within clinically acceptable limits such as ±3° or ±5°.
In accordance with current Technical Guidelines for Clinical Evaluation of Medical Devices, two alternative routes are available for spinal measurement software:
Equivalent Device Comparison Route (Preferred Option): Select this route when comparable marketed products exist with similar core algorithm principles and product differences limited to UI optimization or improved computational efficiency. Note: Comparison shall cover core measurement precision and clinical applicability instead of functional descriptions only.
Clinical Trial Route: Formal clinical trials are mandatory for products adopting innovative algorithm frameworks (novel deep learning architectures) or expanding clinical indications to new spinal disorders, e.g., extending from Adolescent Idiopathic Scoliosis to Early-onset Scoliosis, to validate clinical effectiveness.
In previous reviews of spinal measurement software, the below issues commonly trigger supplementary data requests; project teams are advised to conduct pre-submission self-inspection:
| Common Risk Points Likely to Trigger Supplementary Documentation |
|---|
The user manual claims auxiliary diagnosis while the actual function only covers auxiliary measurement without supporting diagnostic logic.Insufficient volume of training data or monotonous test dataset consisting only of mild scoliosis cases or data from a single device type.Only high accuracy on internal test dataset is provided without consistency analysis versus manual measurements by practicing clinicians.Neglected security for data transmission and storage with no cybersecurity vulnerability scanning performed. |
The registration application of spinal measurement software is a systematic project. The core requirement of review experts lies in ensuring product safety and effectiveness. Enterprises should not merely focus on code operation from a programmer’s perspective, but evaluate whether the generated measurement data is clinically applicable from the perspective of clinicians.
It is recommended that enterprises introduce professional registration and compliance teams at the initial stage of project initiation to clarify product classification (Class II or Class III), and carry out software lifecycle management and cybersecurity verification in parallel. In addition, review authorities encourage applicants to strengthen communication with professional registration teams and regulatory institutions, thoroughly understand relevant regulations and guiding principles, and prepare application documents in a scientific and rigorous manner.
Deda Medical provides full-process supporting services including preliminary strategy judgment, research protocol design, clinical trial assistance and application document integration, covering medical device clinical trials, stability and effectiveness evaluation, registration document compilation and quality system alignment. Only products that integrate algorithm rigor, data compliance and clinical practicality, with impeccable evidence chains constructed in registration dossiers, can stand out in technical reviews and successfully obtain market approval.
