I am writing in regard to the article titled “Validation of the visual body image classification in adolescent idiopathic scoliosis: a retrospective study” by Kim et al. [1]. This study offers an innovative approach to scoliosis screening using visual body images, addressing the critical issue of radiation exposure associated with traditional radiographic methods. However, I would like to highlight certain limitations and areas for further improvement to enhance the robustness and generalizability of the findings.

Firstly, the study’s reliance on a tertiary hospital-based cohort introduces selection bias. The participants, predominantly those already diagnosed with scoliosis or suspected of the condition, may not represent the general population. This bias likely affects the reported sensitivity and specificity of the classification method. Future research should incorporate randomized community-based recruitment to validate these findings across broader demographics. This would not only broaden the demographic representation but also enhance the classification system’s validity across varied patient populations, including those without prior scoliosis screening or diagnosis [2,3].

Secondly, while the authors emphasize the high sensitivity (98.1%) and specificity (98.9%) of the proposed classification method, the notably low sensitivity for type 2 (46.1%) raises concerns. Variability in photography techniques, patient posture, and anatomical differences may contribute to this discrepancy. Further exploration into the integration of artificial intelligence-based image analysis or deep learning tools could offer a solution. Such technologies could minimize subjectivity and improve consistency in identifying subtle deviations in posture associated with type 2 scoliosis [4,5].

Another significant limitation is the exclusion of participants with leg-length discrepancies, skeletal dysplasia, or other conditions that may influence posture. While this was done to ensure methodological clarity, it restricts the applicability of the findings to real-world populations, where such comorbidities are prevalent. Future research should consider expanding participant criteria to include these populations, enabling a more comprehensive evaluation of the classification system. For instance, leg-length discrepancies are known to influence pelvic alignment, which may alter the observed spinal curvature. Including such cases in future studies could provide crucial data on the system’s applicability and robustness in more complex clinical scenarios. Incorporating these variables into future studies would enhance the method’s clinical relevance [6,7]. The study also lacks sufficient details on interobserver variability. Although the reported kappa values indicate good interobserver reliability, the moderate interobserver reliability (κ=0.751) suggests potential challenges in standardizing the classification process among different evaluators. Addressing this variability, possibly through enhanced training or automated analysis, would strengthen the method’s utility in clinical and non-specialist settings [5,8].

Lastly, ethical considerations surrounding the use of photographic data require attention. While informed consent was obtained, anonymization and secure data storage protocols were not explicitly discussed. This omission may raise concerns about participant privacy and compliance with international ethical standards. Best practices include applying robust encryption for data storage, limiting access to authorized personnel, and ensuring compliance with frameworks like GDPR (General Data Protection Regulation) or HIPAA (Health Insurance Portability and Accountability Act). For anonymization, employing advanced techniques such as facial obfuscation or replacing identifiers with pseudonyms could further enhance participant privacy. Previous studies have highlighted how inadequate anonymization can lead to breaches of trust and confidentiality, potentially jeopardizing future research collaborations. These measures are critical for safeguarding participant data and maintaining ethical integrity [8,9].