Intelligent In situ Printing of Multimaterial Bioinks for First‐Aid Wound Care Guided by Eye‐In‐Hand Robot Technology

Background

In situ bioprinting deposits bioinks directly into defects, but single-plane printing with three-degree-of-freedom (3-DOF) systems struggles with large wounds of varying shape and anatomical location. Earlier approaches relied on printing codes predesigned from computer models, micro-computed tomography or 3D scanners, which are difficult to adapt to each patient's defect and require repeated manual calibration. Vision-guided printing with an eye-to-hand camera adapts to targets in real time, but its limited field of view requires multiple coordinate correlations and can leave volumetric regions incompletely modeled.

Design and approach

INSIGHT combines a visual guidance system, a 6-DOF robot arm (EPSON VT6L) and a printhead with computer-controlled pneumatic valves. An Intel RealSense D405 depth camera on the end effector captures color and stereoscopic depth images; wound regions are isolated by color masking and contour detection, converted to point clouds, bounded with alpha shapes, which follow defect outlines more closely than a convex hull, and turned into a robot travel path whose point density is set by line and grid numbers. Because the scanned surface coordinates are used directly, separate coordinate calibration is not needed. Images taken from several angles can be merged to recover regions hidden from a single viewpoint, layers are rescanned after each print, and the arm hovers about 2 cm above the wound between filaments. Multiple syringes on separate valves allow inks with different rheology to be switched by rotating the end effector, and the printhead can extrude or spray.

How it was evaluated

Scanning was characterized by processing time against wound area, field of view against scan height, and area and depth error across 36 color combinations and across scan angles from -75° to 75°. Printing was demonstrated with an alginate, gelatin and cellulose nanofiber (CNF) ink on pattern motifs, a quick response (QR) code puzzle with missing pieces, ex vivo wounds in 300-day-old Sprague-Dawley rats, tilted and multi-angle targets, wound phantoms on curved polylactic acid substrates, and ex vivo porcine wounds from a pig of about 40 kg, including large burn wounds for spray printing, which was also modeled with an air blast atomizer simulation. A photocurable poly(glycerol-co-sebacate) acrylate (PGSA) ink was used to print a covering band-aid over filled wounds. For in vivo testing, microgels of gelatin, methacrylated gelatin (GelMA) and polyethylene oxide (PEO) were printed into two 6 mm full-thickness dorsal wounds in 12-week-old male db/db diabetic mice, crosslinked with UV light for 30 s and covered with Tegaderm, with Tegaderm alone as the comparison and wound closure followed on days 0, 3, 6 and 10.

Key results

Scan processing time increased from 46.3 s for a 1 cm2 area to 172.9 s for 100 cm2, so geometry processing stayed under 3 min. Raising the scan height from 20 to 55 cm enlarged the field of view from 0.05 m2 to about 0.60 m2, although thicknesses of 2 to 3 mm were not detected above 30 cm. Area recognition exceeded 95% accuracy for all 36 color combinations and reached 99% across scan angles from -75° to 75°, while depth recognition showed some deviation, particularly for narrow depths.

The system printed into recognized targets regardless of position, restored the missing pieces of a QR puzzle with squares of about 5 mm × 5 mm, and filled side-located wounds by tilting to ±30°. Merging point clouds captured at 0°, 30° and 40° filled regions that a single 0° capture missed and that otherwise remained unprinted. With rescanning between layers, four printing passes filled a wound cavity to near skin level, and the printed volume matched the empty wound volume of about 3 mL; the storage modulus of the crosslinked ink was close to the 440 to 6620 Pa range of the dermis.

In the ex vivo porcine model, a deep wound of about 10 cm was filled by extrusion in 10 iterations, and spray mode covered a 10 cm × 10 cm area 12 times faster than extrusion. The simulation predicted particles of about 56 μm spread over 18.81 cm2 at a 5 cm printing height and a maximum pressure of 6.75 × 10^-5 Pa at the wound center, while experimentally observed particles ranged from 903 μm to 283 μm between 0.5 cm and 2 cm from the center, which the authors attribute to aggregation not included in the model. A PGSA band-aid printed over a 1.5-fold enlarged mask and cured for 30 s at 800 mW cm−2 covered the filled wound, with 10 psi used for wound filling and 5 psi for the band-aid.

In db/db mice, scanning and path generation took about 30 s per wound and printing about 30 s. Both groups exceeded 50% wound closure on day 6, with a slight difference in healing speed on day 10. Masson's trichrome staining showed well-developed granulation tissue and higher cell density in microgel-treated wounds, and the area of hyperproliferative neo-epidermis was significantly larger than with Tegaderm alone (p < 0.05).

Open questions

The authors note that limitations in the acquired data can reduce the accuracy of scene reconstruction, and that the current color detection may miss wound edges or fail to distinguish wounds from blood flow, so primary treatment such as blood suction appears necessary before clinical use. They propose advanced convolutional neural techniques to improve image fidelity for planning and are developing microgel bioinks containing human mesenchymal stem cells and oxygen-generating microparticles for diabetic wounds. The depth camera did not resolve 2 to 3 mm thicknesses at scan heights above 30 cm.

How it connects to other research

A mobile skin bioprinting system (Albanna et al., Scientific Reports 2019) used integrated imaging to deliver layered autologous or allogeneic fibroblasts and keratinocytes into wounds, whereas INSIGHT moves an eye-in-hand depth camera with a 6-DOF arm to rescan wounds from multiple viewpoints and switches among acellular inks and between extrusion and spray modes.