<p>Prosthetic selection critically influences rehabilitation outcomes for lower-limb amputees, yet conventional approaches often rely on subjective clinical judgment and static protocols, frequently overlooking individualized patient factors. This study presents ProsthetiX-AI, a clinical decision support system that integrates a deterministic policy engine with evidence-based reasoning, supported by an explanation module from large language models, to deliver personalized prosthetic recommendations. The framework dynamically analyzes patient-specific parameters, such as amputation level, mobility classification, comorbidities, weight, and biomechanical characteristics to generate recommendations aligned with established clinical guidelines. A core innovation of the system lies in its ability to transparently justify outputs by retrieving peer-reviewed evidence, including mobility classification standards and weight-based component selection criteria, thereby enhancing interpretability for clinicians. Delivered through an interactive web interface, the system supports automated reporting, safety validation for weight-based compatibility, and real-time monitoring. Evaluations across transtibial and transfemoral amputations, as well as complex comorbidity profiles, demonstrated sub-millisecond latency, reliable multi-user handling, and adherence to validated recommendations. Quantitative evaluation showed an accuracy of 0.72–0.89 and Cohen’s <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$\kappa =0.64$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>κ</mi> <mo>=</mo> <mn>0.64</mn> </mrow> </math></EquationSource> </InlineEquation>–0.85, achieving agreement within the range of clinician–clinician variability (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">$\kappa =0.57$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>κ</mi> <mo>=</mo> <mn>0.57</mn> </mrow> </math></EquationSource> </InlineEquation>). Feedback from five clinicians and five prosthetic users reported high ratings for accuracy (4.76/5) and usability (4.54/5), highlighting the system’s clinical potential. Despite modest, vignette-based samples, results suggest that citation-linked recommendations can augment clinical decision-making. By emphasizing transparency, scalability, and clinician oversight, ProsthetiX-AI addresses challenges in adopting artificial intelligence in healthcare and provides a foundation for patient-centric decision support in resource-constrained environments.</p>

ProsthetiX-AI: An LLM-based clinical decision support system for evidence-based prosthetic recommendations

期刊论文
发表日期：

2026年4月11日

查看全文

Vidyapati Kumar,
Dilip Kumar Pratihar

摘要

Prosthetic selection critically influences rehabilitation outcomes for lower-limb amputees, yet conventional approaches often rely on subjective clinical judgment and static protocols, frequently overlooking individualized patient factors. This study presents ProsthetiX-AI, a clinical decision support system that integrates a deterministic policy engine with evidence-based reasoning, supported by an explanation module from large language models, to deliver personalized prosthetic recommendations. The framework dynamically analyzes patient-specific parameters, such as amputation level, mobility classification, comorbidities, weight, and biomechanical characteristics to generate recommendations aligned with established clinical guidelines. A core innovation of the system lies in its ability to transparently justify outputs by retrieving peer-reviewed evidence, including mobility classification standards and weight-based component selection criteria, thereby enhancing interpretability for clinicians. Delivered through an interactive web interface, the system supports automated reporting, safety validation for weight-based compatibility, and real-time monitoring. Evaluations across transtibial and transfemoral amputations, as well as complex comorbidity profiles, demonstrated sub-millisecond latency, reliable multi-user handling, and adherence to validated recommendations. Quantitative evaluation showed an accuracy of 0.72–0.89 and Cohen’s $\kappa =0.64$ $κ = 0.64$ –0.85, achieving agreement within the range of clinician–clinician variability ( $\kappa =0.57$ $κ = 0.57$ ). Feedback from five clinicians and five prosthetic users reported high ratings for accuracy (4.76/5) and usability (4.54/5), highlighting the system’s clinical potential. Despite modest, vignette-based samples, results suggest that citation-linked recommendations can augment clinical decision-making. By emphasizing transparency, scalability, and clinician oversight, ProsthetiX-AI addresses challenges in adopting artificial intelligence in healthcare and provides a foundation for patient-centric decision support in resource-constrained environments.