AI-Enhanced Multimodal Sensing for Early Detection of Diabetic Foot Ulcers: A Cost-Effective Wearable Solution

Abstract

This paper presents a novel wearable IoT system integrating multi-modal plantar pressure mapping and thermal imaging for early detection of diabetic foot ulcers (DFUs). Our solution combines six FlexiForce A502 pressure sensors (0-500 kPa range) and four MLX90614 infrared temperature sensors (±0.5?C accuracy) with a hybrid CNN-BiLSTM deep learning architecture, achieving 94.2% prediction accuracy on a clinical dataset of 75 diabetic patients over 90 days. The system demon- stratessignificantimprovementsoverexistingapproaches,reduc- ing false positives by 32% compared to single-modality systems whilemaintainingaproductioncostof$14.80perunit.Real-time edge processing on an ESP32-C6 microcontroller enables 28 ms inference latency, with piezoelectric energy harvesting extending operational lifetime to 72 hours. Clinical validation shows 40% reductioninhospitalizationratesthroughearlyintervention,with potentialannualhealthcaresavingsexceeding$9billionintheUS alone.Thisworkbridgesthegapbetweenclinic-gradediagnostics and community healthcare through explainable AI and cost- effective wearable technology.

Introduction

1. Clinical Context

• Diabetes affects over 537 million adults worldwide.

• 34% develop diabetic foot complications; DFUs are the leading cause of diabetes-related hospitalizations and 20% can result in amputation.

• Existing diagnostic tools like MRI and pressure mats are effective but costly and inaccessible in low-resource settings.

2. Limitations of Current Technologies

• Pressure-only sensors miss early inflammation.

• Temperature-based systems lack localization.

• Vibration alerts detect ulcers too late.

3. Proposed Innovations

• Multimodal Sensor Fusion: Combines pressure (100 Hz) and bilateral foot temperature (0.1°C resolution).

• Edge-optimized CNN-BiLSTM AI: Processes spatial-temporal patterns with 94.2% accuracy on low-power embedded hardware (ESP32).

• Clinical Decision Support: Integrates with hospital EHRs (HL7/FHIR) for real-time risk assessment.

4. Literature Insights

• Prior solutions lacked either:

  • Multimodal sensing,

  • Advanced AI for sequential data, or

  • Real-time/edge computing support.

• The proposed hybrid CNN-BiLSTM model addresses these gaps by offering accurate, low-latency, interpretable diagnostics.

5. System Design

• Sensors: FlexiForce (pressure) + MLX90614 (temperature).

• Processor: ESP32-C6, 160 MHz, 320 KB SRAM.

• Power: LiPo battery with piezoelectric energy harvesting.

• AI Model: CNN-BiLSTM with attention, trained for 3-class classification (low, medium, high risk).

6. Experimental Results

• Accuracy: 94.2%

• Sensitivity: 92.4%

• Specificity: 95.1%

• F1-Score: 0.93

• Comparison: Outperformed previous models (Bai et al., Berugu et al., and S?ahin & Cingil).

7. Clinical Impact

• 88% of high-risk patients correctly identified within 90 days.

• 40% reduction in hospitalizations (from 18.7 to 11.2 days/year).

• Cost: ~$0.23/day per patient vs. $3,120/year for standard care.

Conclusion

Thisresearchdemonstratesthefeasibilityandimpactof an AI-enhanced, multimodal wearable system for early de- tection of diabetic foot ulcers (DFUs). By integrating plantar pressure mapping and thermal imaging with a hybrid CNN- BiLSTM deep learning model, the proposed solution achieves a high prediction accuracy of 94.2% and significantly reduces false positives compared to single-modality systems. The system’s real-time edge processing, low production cost, and energy harvesting capabilities make it practical for continu- ous, community-based monitoring and scalable deployment in resource-limited settings. The clinical validation on 75 diabetic patients over 90 days highlightsseveralkeybenefits:a40%reductioninhospitaliza- tionrates,earlierriskstratification,andsubstantialcostsavings compared to conventional MRI-based screening. The explain- able AI framework, with Grad-CAM visualizations, not only enhancesclinicaltrustbutalsosupportstargetedinterventions, potentially preventing ulcer progression and amputations. Despite these strengths, some limitations remain. Sensor drift requires regular calibration, and the current study cohort hadlimitedrepresentationofneuropathicpatients.Addressing these will be a focus of future work, alongside the planned integration of photoplethysmography (PPG) for blood flow assessment, federated learning for privacy-preserving model updates, and expansion to multi-center clinical trials. Insummary,thisworkbridgesthegapbetweenhospitalgradediagnosticsandaccessiblepreventivecarefordiabeticfootcomplications.Theplatform’sflexibility,accuracy,and cost-effectiveness position it as a promising tool for reducing theglobalburdenofDFUs.Futuredirectionswillinclude regulatory validation, integration with smart bandage systems forclosed-loopcare,andopen-sourcereleaseofcodeanddatasetstoacceleratefurtherresearchandreal-worldadoption. ThisworkdemonstratesthatedgeAIenabledmultimodalwearablescanachievehospitalgradeDFUpredictionaccuracyatcommunityhealthcarecosts.Ourclinicalvalidationwith75patientsover90daysconfirmsthesystem’spotentialtotransformdiabeticfootcarethrough: • Early detection of pre-ulcerative states (14 days median lead time) • PersonalizedriskstratificationusingexplainableAI • Seamlessintegrationwithexistinghealthcareinfrastruc- ture Future work will focus on multi-center trials (n=500+) and regulatory approval pathways for clinical deployment.

References

[1] InternationalDiabetesFederation,“DiabetesAtlas,10thed.,”2023. [2] D.M.Baietal.,“IoT-BasedFootUlcerDetection,”IEEESens.J.,vol. 21, no. 5, 2023. [3] M.S¸ahinandD.Cingil,“VibrationAlertsinDFUMonitoring,”J.Med.Devices, vol. 15, no. 2, 2023. [4] S. Berugu et al., “IoT/ML for DFU Detection,” MATEC Web Conf.,vol. 392, 2024. [5] L. J. Joudeh et al., “Smart diabetic foot ulcer scoring system (ScoreD-FUNet),”ScientificReports,vol.14,2024.[Online].Available:https: //www.nature. com/articles/s41598-024-62076-1 [6] R.J.Hinchliffeetal.,“AMulti-FacetedDigitalHealthSolutionforMon-itoring and Managing Diabetic Foot Ulcers,” J Diabetes Sci Technol.,2024. [Online]. Available: https://pubmed.ncbi.nlm.nih.gov/38732781/ [7] A.S.Surietal.,“Theroleofmachinelearninginadvancingdi-abetic foot: a review,” Frontiers in Endocrinology, vol. 15, 2024.[Online].Available:https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2024.1325434/full [8] A.K.Srivastavaetal.,“DFUCare:deeplearningplatformfordi-abetic foot ulcer detection, analysis, and monitoring,” Frontiers inEndocrinology, 2024. [Online]. Available: https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2024.1386613/full [9] S.S.Pateletal.,“MultimodalIdentificationofMolecularFactorsLinkedtoSevereDiabeticFootUlcers,”FrontiersinEndocrinology,2024.[On-line]. Available: https://pmc.ncbi.nlm.nih.gov/articles/PMC11476782/ [10] J. M. Najafi et al., “A Review of Wearable Sensor Systems to MonitorPlantar Pressure and Shear Stress in the Foot,” J Diabetes Sci Technol.,2019. [Online]. Available: https://pubmed.ncbi.nlm.nih.gov/31899409/

Copyright

Copyright © 2025 S Vamshi Kashyap, Rathan Swaroop B, Dr. Manoj Kumar M. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.