Technical Feasibility of a Multimodal Electro-Optical and Infrared Imaging Pipeline for Postoperative Wound Assessment After Total Knee Arthroplasty: Formative Evaluation Study (Preprint)
Preprint
- 28 April 2026
- preprint
- Published by JMIR Publications Inc. in JMIR Preprints
Abstract
BACKGROUND Early detection and monitoring of wound-related complications after total knee arthroplasty (TKA) are critical to optimize outcomes and preserve implants. Conventional postoperative monitoring relies on in-person assessment and patient-reported symptoms, which may delay recognition. Noncontact imaging acquired in the clinic or at home offers a potential route to earlier and more consistent assessment. OBJECTIVE This study aimed to evaluate the technical feasibility and performance of a multimodal AI system integrating electro-optical (EO) and infrared imaging for postoperative wound assessment following TKA and to explore clinical factors associated with early complications. METHODS We conducted a single-center prospective cohort study of 749 patients undergoing primary TKA. Patients underwent standardized multimodal data collection at scheduled postoperative visits. EO-based models localized surgical landmarks and segmented complication-oriented tissue states; a fine-tuned vision-language model provided protocol verification and contextual visual reasoning. EO landmarks mapped to thermal space via deterministic alignment, followed by infrared region-of-interest segmentation and thermal classification. Each module was evaluated on a defined held-out set. The thermal classifier was assessed out-of-fold under 5-fold cross-validation with folds grouped by patient, in knees imaged on postoperative day 14 or later. Discrimination for clinician-adjudicated infection-related complications served as the primary outcome. Comorbidities and baseline laboratory values were explored for associations with complications. RESULTS Among 749 patients, 26 (3.5%) had a clinician-adjudicated infection-related complication. On held-out evaluation, the EO landmark detector reached a sensitivity of 0.967 (95% CI 0.954-0.979) and positive predictive value of 0.955 (95% CI 0.940-0.970), and the wound-tissue segmentation model reached a sensitivity of 0.845 (95% CI 0.833-0.857) and positive predictive value of 0.759 (95% CI 0.744-0.772). Infrared region-of-interest extraction isolated the surgical field with a mean intersection-over-union of 0.985 (SD 0.008; 95% CI 0.984-0.986) across 178 held-out patients. The trained thermal classifier discriminated infected scans with an area under the receiver operating characteristic curve (AUROC) of 0.978 (95% CI 0.959-0.992) in out-of-fold, patient-grouped cross-validation. At a representative operating point, sensitivity was 0.808 (95% CI 0.65-0.92), specificity 0.974 (95% CI 0.96-0.99), positive predictive value 0.618 (95% CI 0.49-0.78), and negative predictive value 0.990 (95% CI 0.98-1.00). Exploratory analyses showed nominal associations of chronic kidney disease and diabetes mellitus with postoperative complications that did not remain significant after multiplicity correction. CONCLUSIONS In this formative study, a multimodal EO and infrared imaging pipeline demonstrated technical feasibility for postoperative wound assessment following TKA, providing noncontact discrimination of infection-related complications that compared favorably with established adjuncts. This work provides proof-of-concept for multimodal imaging-based assessment and defines the technical requirements for the prospective, multicenter external validation needed to establish clinical utility.Keywords
All Related Versions
- Published version: JMIR Formative Research, 10, e99244.
This publication has 27 references indexed in Scilit:
- Current Rates and Trends of Venous Thromboembolism After Total Hip and Knee Arthroplasty: An Updated Analysis Utilizing the NSQIP DatabaseArthroplasty Today, 2025
- Efficacy of Debridement, Antibiotics, and Implant Retention in Total Hip and Knee Arthroplasty: A Systematic Review and Meta-AnalysisThe Journal of Arthroplasty, 2025
- Reporting Guidelines for Artificial Intelligence Use in Orthopaedic Surgery ResearchThe Journal of Arthroplasty, 2025
- Electronic Patient-Reported Symptoms After Ambulatory Cancer SurgeryJAMA Surgery, 2024
- Direct Hospital Costs per Case of Periprosthetic Hip and Knee Joint Infections in Europe — A Systematic ReviewThe Journal of Arthroplasty, 2024
- Effect of Smartphone App Postoperative Home Monitoring After Oncologic Surgery on Quality of RecoveryJAMA Surgery, 2023
- Risk stratification in primary total joint arthroplasty: the current state of knowledgeArthroplasty Today, 2019
- Persistent Pulmonary Embolism Rates Following Total Knee Arthroplasty Even With Prophylactic AnticoagulantsThe Journal of Arthroplasty, 2017
- Medical applications of infrared thermography: A reviewInfrared Physics & Technology, 2012
- Characterization of a cDNA for human glutathione-insulin transhydrogenase (protein-disulfide isomerase / oxidoreductase)Biochimica et Biophysica Acta (BBA) - Gene Structure and Expression, 1988