AI-Driven Smart Bandage Shows Promise in Accelerating Wound Healing

Scientists from the University of California and collaborating institutions have developed a-Heal—a closed-loop system that continuously monitors wound status, interprets healing stages using machine learning, and delivers bioelectronic therapies accordingly.

At the core of the a-Heal platform is a wearable patch embedded with a miniature camera and bioelectronic actuators. The device captures wound images every two hours and transmits them wirelessly to an AI system dubbed the “ML Physician.” This algorithm analyzes the wound’s stage—hemostasis, inflammation, proliferation, or maturation—and calculates an optimal treatment trajectory using a model rooted in deep reinforcement learning.

The wearable patch then delivers therapy via electric fields or iontophoretically administered fluoxetine (an antidepressant repurposed for its wound healing properties), adapting treatment in response to changes in the wound's condition. The system’s graphical user interface allows physicians to oversee or override decisions in real time, ensuring a hybrid of automation and clinical oversight.

In a 22-day study using a porcine excisional wound model—a standard proxy for human skin—a-Heal was applied to full-thickness wounds for the first seven days, with follow-up monitoring thereafter. The AI-controlled device dynamically adjusted the timing and dosage of electric field and fluoxetine treatments based on wound stage transitions. Notably, the ML algorithm delayed fluoxetine delivery until the inflammatory phase peaked, maximizing its effectiveness during the proliferative phase of healing.

Treated wounds showed faster closure rates, with one experiment achieving 100% re-epithelialization by day 22. Histological analysis confirmed enhanced epidermal thickness, a 26% reduction in granulation tissue, and a more mature collagen profile in treated wounds. Inflammatory markers such as IL1B were reduced by 61%, while anti-inflammatory genes like IL10 and regenerative factors such as TGF-β1 were elevated.

Further supporting the system’s efficacy, tissue analysis revealed localized fluoxetine delivery with no systemic accumulation, reducing the risk of off-target effects. In contrast, two wounds where treatment was interrupted due to device failure healed significantly more slowly, emphasizing the importance of continuous algorithm-guided care.

While the study’s scope was limited—only a handful of wounds were treated, and the therapies used are not yet FDA-approved for wound healing—it lays the groundwork for future trials. 

a-Heal’s compact, wireless design and AI-guided therapy could extend specialized wound care to patients lacking consistent clinical access. The system’s modular architecture also opens the door to expanding therapeutic options, including antimicrobial delivery for infected wounds—an avenue the team plans to explore next.