On‐Patient Temporary Medical Record for Accurate, Time‐Sensitive Information at the Point of Care

Accurate medical recordkeeping is important for personal and public health. Conventional forms of on‐patient medical information, such as medical alert bracelets or finger‐markings, may compromise patient privacy because they are readily visible to other people. Here, the development of an invisible, temporary, and easily deployable on‐patient medical recordkeeping system is reported. Information is stored in unique patterns of spatially distributed near‐infrared (NIR) fluorescent quantum dots (QDs), which are delivered to the skin using dissolvable microneedle arrays. The patterns are invisible to the naked eye but detectable with an infrared camera, which can extract information with >98% accuracy using automated pattern recognition software. By encapsulating NIR QDs in an FDA‐approved biodegradable polymer, biodegradation rates can be tuned so that the encoded medical information can be conveyed in both a spatial and temporal manner, with some components fading within 100 days and others persisting for 6 months. This may be particularly useful for administering a series of vaccinations or treatments by indicating if enough time has passed for the patient to receive the next dose. Importantly, this system contains no personal information, does not require connection to a centralized database, and is not visible to the naked eye, ensuring patient privacy.


Introduction
[11] However, the reality of many global settings is that medical records are often unavailable, for example, due to misplacement or lack of immediate access to paper-based records, or The on-patient recordkeeping system is composed of three parts: 1) the microneedle array, 2) the invisible, fluorescent QD dye, and 3) the imaging hardware/software.[29] For these reasons, a dissolvable microneedle array was selected as a means for on-patient medical recordkeeping.
Controlling the spatial distribution of the NIR QD dyes loaded in microneedles essentially enables information encoding.The QDs must absorb and emit light between 850-1100 nm to minimize absorption by tissue and be invisible to the naked eye (i.e., must be longer than the wavelength of the upper limit of visible light, > 750 nm).The QDs composed of a copperindium-selenide core and an aluminum-doped zinc sulfide shell (≈4 nm size) [17] with oleic acid surface functionalization were prepared (Figure 2A) to emit light between 800-1000 nm with a photoluminescence intensity peak at 914 nm (Figure 2B) and a photoluminescence quantum yield of 40.9 ± 0.2% (Figure S1, Supporting Information).The QDs do not contain lead or cadmium, have been reported to be non-toxic against RAW 264.7 mouse macrophage cells, even at high concentrations (1000 μg mL -1 ), and retain ≈ 15% of their initial signal intensity after 5 years of simulated sun exposure. [17]In this report, we purchased commercially available negative PDMS microneedle molds to make QD-loaded microneedle patches (Figure 2C), which feature 10 × 10 square-based pyramid needles with a height of 800 μm, base of 200 μm, and pitch of 500 μm (Figure 2D).Before directly loading these QDs, however, in order to ensure the nanoscale QDs are not rapidly cleared by the body, they must be encapsulated within polymer microparticles. [17]canning electron microscopy (SEM) (Figure 2E) and Coulter Counter analysis (Figure S2, Supporting Information) of the QDs following encapsulation with poly(methyl methacrylate) (PMMA), an FDA-approved non-biodegradable polymer, revealed spherical particles (QD-PMMA) with an average size of 9.5 ± 8 μm (Figure 2A).Encapsulation had little effect on the emission wavelength of the QDs with a minor shift of the peak (Figure 2B).QD-PMMA-loaded microneedles were loaded to negative microneedle mold using a centrifugation technique to concentrate the QD-PMMA microparticles at the needle tips.The negative mold was then filled with PVA:sucrose solution to form the body of the microneedles (Figure S3,Supporting Information).After drying, the microneedle arrays composed of non-toxic materials safe for human use were ready for use (Figure 2F).The pattern can be read with a handheld imaging device to retrieve information.B) The microneedle array can be modified to achieve a change in the pattern after a certain amount of time by the selective loading of biodegradable fluorescent microparticles (as indicated by the green dot).This is used to encode information in both a spatial and temporal manner.
In order for this system to be accessible in austere settings or be rapidly deployed during emergencies such as infectious disease outbreaks, the imaging components must be easy to use and transport, inexpensive, and reliable.Accordingly, we developed a robust and low-cost NIR imaging device composed of a handheld reader containing a ring of 780 nm LEDs for fluorescent illumination which connects to a microprocessor for realtime imaging (Figure S3, Supporting Information), analogous to current contactless IR thermometers. [30]This handheld imaging device that reads patterns of fluorescent QDs in situ (Figure 2D) was designed as a standalone unit that does not require internet connectivity or uploading of patient data to a centralized database to add an extra measure of patient privacy, data security, and applicability in remote settings.

Long-Lasting, Invisible On-Patient Recordkeeping
We first examined the application of QD-PMMA-loaded microneedle array patches in ex vivo pigmented human skin (n = 5) to ensure the applicability of the technology in diverse populations with a range of skin pigmentations.To confirm proper skin penetration without the microneedles breaking or bending, a mechanical compression test was first performed on 10 × 10 microneedle patches (n = 5) and on individual microneedles (n = 4) (Figure S4, Supporting Information).The compression load test demonstrated that the maximum force a microneedle patch or a single microneedle can endure without breaking exceeded the minimum force required to puncture human skin, and therefore, our platform is a viable method for QD delivery.To apply patches with a consistent application force, a spring-loaded applicator was used.Following the application for 2 min, imaging with the handheld device revealed the successful transfer of the fluorescent particles, while no pattern was visible to the naked eye (Figure S3, Supporting Information).
After achieving successful microneedle applications, a preliminary information encoding scheme development followed.In-formation encoding must be simple, secure, and versatile.One method that satisfies these criteria is the select removal of 2 × 2 squares of microneedle pins within a 10 × 10 array (Figure 2G).The unintentional loss of 2 × 2 squares during application is highly unlikely, but the removal of four pins provides 4× redundancy to account for the potential loss of random, single pins during application.The removal of select pins is straightforward (and could potentially be automated) and easily convertible to binary for automatic pattern generation and information encoding/retrieval (Figure S5, Supporting Information).Therefore, we selected this 2 × 2 square removal method as the encoding scheme.
The encoding scheme has three additional design aspects.First, the 2 × 2 square in the top right corner of the microneedle array is always removed for pattern alignment and desymmetrization.Second, the outer edge of the microneedle array is left intact.Third, 1-4 squares of microneedles are removed such that the removed areas do not overlap.With these limitations, fifteen 2 × 2 squares are available for information encoding that can be used to generate 32 767 unique patterns, each pattern potentially corresponding to distinctive medical record information.The number of patterns can be tuned further by increasing the patch size and the number of needles changing the number of pins per block and redundancy.
To examine the encoding scheme and lifetime of the transferred patterns, three encoded QD-PMMA microneedle arrays (PMMA 1-3, Figure 2H) were applied to Wistar rats (n = 3) (Figure S6, Supporting Information).Light microscopy imaging of the microneedle arrays after 2 min application revealed dissolution of the microneedle tips, with deposition of the QD-PMMA microparticles in the rat skin (Figure S7, Supporting Information).Analysis of the transferred patterns revealed that no more than 5 microneedles were lost upon application, indicating a > 93 (±3) % transfer efficiency, which is well within the redundancy limits designed into the platform (Figure 2H).Over the first month, the intensity of the transferred patterns decreased, most likely due to the loss of QD-PMMA particles deposited near the surface of the skin (Figure 2H).After a month, the patterns appeared to stabilize (Figure S8, Supporting Information), though it is likely that the patterns continue to fade but relatively slowly due to unpreventable minor biological clearance or photobleaching reasons.The initial decrease in intensity followed by a region of very slow decline agrees with previous reports. [17] this proof-of-concept PMMA platform was not biodegradable, the patterns are expected to remain stable and able to encode information for a long duration on the order of years.This study demonstrated imaging out to 6 months, during which time the patterns are clearly visible (Figure 2H).Because future clinical development will focus on temporary biodegradable systems, longitudinal imaging was stopped after 6 months.We next aimed to prepare short-lived patterns and examine the potential for automated pattern recognition.

Shape-Changing Patterns for Time-Dependent Information
On-patient recordkeeping can offer benefits for different durations of time, depending on the application.[33] Accordingly, we designed a biodegradable system in which the signal disappears at a prescribed time after administration.
We hypothesized that by encapsulating the QDs in a biodegradable polymer, the transferred patterns would remain visible until the encapsulating polymer microparticles degrade, leaving the nanoscale QDs to be cleared by the body. [34,35]To test this, QDs were encapsulated in poly(lactic-co-glycolic acid) (PLGA), an FDA-approved, and widely used biodegradable polymer.PLGA degrades by hydrolysis and the rate of degradation can be tuned by changing the molecular weight of the polymer, the lactic acid (LA) to glycolic acid (GA) ratio, and the chain end functionality. [36]Encapsulation of the QDs within PLGA (QD-PLGA) was not found to affect the properties of the QDs.The photoluminescence (peak emission at 932 nm) (Figure 3A) and quantum yield (8.7%) (Figure S1, Supporting Information) were very similar to QD-PMMA, and the size of the QD-PLGA microparticles was found to be 6.6 ± 6 μm (Figure S2, Supporting Information).
A preliminary in vitro degradation study was performed to provide an estimate of the lifetime of the QD-PLGA microparticles in PBS at 37 °C.The QD-PLGA microparticles were imaged via SEM to examine the rate of particle degradation (Figure 3B).At time 0, the QD-PLGA microparticles were well-formed and spherical in shape.The QD-PLGA microparticles showed minimal degradation after 16 days, significant degradation at day 31 (Figure S9, Supporting Information), and by day 41, no particles could be observed (Figure 3B).The signal intensities of these microparticles were also examined in glucose-, ethanol-, and collagenase-mixed PBS solutions and in ex vivo human skin to study their signal retention in the presence of inorganic ions, sugar, alcohol, and enzymes in the skin.When the QD brightness was quantified for 7 days using Image J (National Institutes of Health), there was no significant reduction in signal intensity in all of the experimental groups (Figure S10, Supporting Information).The biocompatibility of QD-PMMA and QD-PLGA microparticles was also assessed in vitro using HeLa cells, and the results indicated that the microparticles are not cytotoxic (Figure S11, Supporting Information).
To examine the rate of microparticle degradation in vivo and confirm whether the transferred patterns fade and eventually disappear, QD-PLGA microparticles were loaded into microneedle arrays and applied to Wistar rats (n = 4).Again, only a very small number (2-3) of randomly located microneedle pins were lost during application and the loss of an unintentional 2 × 2 square of microneedles was never observed.Imaging of QD-PLGA patterns revealed the pattern could be imaged in vivo for 118 days (Figure 3C), with signal loss and pattern fading occurring between days 100-118.This is significantly shorter than the non-biodegradable QD-PMMA patterns (> 6 months) but more than double the QD-PLGA microparticles incubated in vitro (41 days).The longer lifetime of the QD-PLGA patterns is potentially due to the relatively lower moisture content of the skin compared to in vitro conditions which slows the rate of hydrolysis.Once free, it is likely that the QDs are cleared via renal pathways, as has been shown for particles of a similar size in prior studies. [34,35]n addition to the fully short-term medical record system, hybrid patterns with long-lasting sections (QD-PMMA) and rapidlyfading sections (QD-PLGA) were designed to demonstrate shapechanging patterns, for example, to indicate that a minimum time has passed and a patient is ready for a next dose of medication or vaccine.To demonstrate these shape-changing patterns, two hybrid designs were prepared (Figure 3D).The first design demonstrates a simple change in shape from a square to a cross.The second design is composed of an encoded top section and a degradable lower section.The latter design allows for variable encoding in the permanent top section with the lower, temporary section providing the relative indication of time.These were applied to rats, and imaged to reveal successful transfers of both primary patterns.After 4 months, the rats were imaged again, and the temporary sections had faded, leaving the new, secondary patterns that can be clearly distinguished from the primary pattern (Figure 3D).The change in shape reflects the loss of the temporary QD-PLGA sections within the pattern.This is the first report of a temporally encoded, shape-changing on-patient recordkeeping and may open a wide range of new applications for this technology.The temporary patterns reported lasted ≈118 days.To accommodate longer intervals, for example, to record the annual administration of medications or vac-cines, encapsulating the QDs in a more hydrolytically stable PLGA (or other biodegradable polymers such as poly(lactic acid) or poly(caprolactone)) may extend the lifetime of the fading pattern. [36,37]

Automated Pattern Identification
For reliable field use, pattern identification and information retrieval can be automated using computer algorithms to avoid human error.We demonstrated pattern read-out using a pattern recognition algorithm based on the AlexNet neural network image classifier. [38]The modified AlexNet was trained using a transfer-based learning approach to minimize the computational costs.In the transfer-based machine learning approach, a pre-existing convolutional neural network is established with all layers similar to the pre-existing classifier except the last layer fine-tuned to specific patterns of interest. [17]This approach has been shown to be a robust method for microneedle-based pattern detection. [17]A custom MATLAB script was also developed to generate synthetic patterns in any arbitrary array for training and validation.Images generated for each pattern were adjusted (e.g., rotation, translation, brightness, noise, etc.) to minimize location-or noise-induced error in classification (Figure S12, Supporting Information).Two classification convolutional neural networks (CNNs) were trained independently to detect the set of patterns used in PMMA or PLGA particles (Figure S13, Supporting Information).
Imaging these patterns revealed some variations in intensity between individual NIR fluorescent pins, which may be caused by inconsistent dye loading into the microneedles and heterogenous rat skin.Although the brightness of pins initially decreased, the spatial distribution of the pins within patterns was retained, and each unique pattern remained clearly discernible.The automated pattern recognition algorithm was successful in distinguishing and identifying three unique QD-PMMA patterns with over 98% probability for the entire duration of the study.Here, the probability of classification is defined as the percentage confidence of the image classifier algorithm that a given image matches a pattern output by the algorithm as opposed to other patterns. [17]Importantly, pattern identification was successful even when pin intensities faded over time, indicating the proof of concept for redundancy in the encoding scheme.
Following the success of the pattern recognition for PMMA1-3, four PLGA patterns (PLGA1-4) were analyzed using the neural network to examine the accuracy of the pattern recognition over time as the PLGA1-4 patterns fade (Figure 4A).The PLGA 1-4 patterns successfully faded at approximately the same rate and were identifiable to a high probability for as long as the patterns were visible (100 days) (Figure 4B).Beyond 100 days, the probability of a correct identification decreased dramatically.The low probability at Day 118 reflects the loss in clarity and intensity of the PLGA patterns which, by this stage, could no longer be accurately identified by the algorithm.Considering this, the identifiable lifetime of PLGA patterns was determined to be ≈100 days, after which, small sections or individual pins within the pattern may still be slightly visible, but the encoded information is no longer retrievable.

Conclusion
A versatile platform for accurate, invisible on-patient recordkeeping has been developed in this study.The transfer of long-lasting (> 6 months), temporary (100 days), and shape-changing patterns was achieved using a dissolvable microneedle array embedded with fluorescent microparticles.This non-permanent and temporary on-patient recordkeeping provides a means of reliable and discrete temporal (i.e., weeks to months) and spatial (i.e., 32767 unique patterns) information storage useful not only in time-sensitive public health control circumstances, but also for medical diagnostics, biosensing, and disease management.The scope of application can potentially extend to non-medical incident markings at mass gatherings, analogous to stamping at sporting events, festivals, and conventions.
QD Synthesis: The QD synthesis was performed according to a previously reported method. [17]To form the CuInSe core: Copper (I) iodide (CuI, 0.286 g, 1.5 mmol) and indium (III) acetate (In(OAc) 3 , 0.438 g, 1.5 mmol) were added to a mixture of 1-dodecanethiol (1.5 mL) and 1octadecene (30 mL).The mixture was degassed under vacuum for 20 min, purged with nitrogen for 20 min, then heated to 120 °C and degassed for a further 45 min.Oleic acid (1.5 mL) was then quickly added via a syringe and the solution was degassed for 20 min and purged with nitrogen for 20 min.The solution was then heated to 175 °C before the injection of the selenium stock solution.To prepare the selenium stock solution, selenium (0.237 g, 3 mmol) and oleylamine (3 mL) were mixed with 1-dodecanethiol (3 mL) and degassed for 30 min at 60 °C.Once prepared, the selenium stock was injected into the main reaction mixture which was then heated to 200 °C.The solution was kept at 200 °C for 30 min under a nitrogen atmosphere.
To the CuInSe core solution, a zinc/aluminum shelling solution was added which was prepared as follows.Zinc acetate (Zn(OAc) 2 , 5.5 g, 30 mmol) was added to a solution of oleylamine (30 mL) and 1-octadecene (30 mL), degassed under vacuum for 20 min and purged with nitrogen for 20 min.Following this, the solution was heated to 120 °C and degassed under vacuum.Separately, aluminum isopropoxide (Al(IPA) 3 , 1.83 g, 9 mmol) was added to a solution of dodecanethiol (5.4 mL) and 1octadecene (36 mL) and degassed for 20 min under vacuum then purged with nitrogen for 20 min.Following this, the solution was sonicated at 60 °C for 60 min.The zinc and aluminum solutions were then mixed and added to the CuInSe core solution at a rate of 0.5 mL min -1 .At the same time, dodecanethiol (15 mL) was added to the CuInSe core solution at a rate of 0.1 mL min -1 .The reaction was allowed to continue for 5 h to complete the shelling reaction.Once cooled, the QDs were purified by precipitating twice into acetone, once into a 50:50 acetone:methanol mixture, and twice more into methanol.Between each precipitation, a minimal volume of toluene and oleic acid was added to resuspend the QDs and ensure correct surface coating and solubility.The purified QDs were then dried and stored in the dark until use.
QD Encapsulation: QD encapsulation was performed following an emulsion/solvent evaporation technique. [17]QDs (150 mg) and either PMMA, PLGA1, or PLGA2 (100 mg) were dissolved in DCM (2 mL).The QD solution was then added to a cold 1 wt% PVA (88% hydrolyzed, 31 kDa) solution (20 mL) and emulsified at 12 000 rpm for 1 min (T 18 digital ULTRA-TURRAX homogenizer [IKA Works]).The resulting emulsion was immediately poured into a separate 1 wt% PVA solution (30 mL) and stirred at 250 rpm overnight to allow the evaporation of the DCM.Following this, the solution was poured into a 50 mL centrifuge tube and centrifuged at 2000 rpm to collect the microparticles.The filtrate was discarded and the microparticles were washed four times by adding deionized water followed by centrifugation.Finally, the particles were resuspended in a small amount of water and filtered through a 75 μm filter to remove large aggregates.The QD-microparticles were then dried and kept in the dark until use.
Photoluminescence Spectroscopy: Photoluminescence emission spectra were measured using a thermoelectrically-cooled silicon camera (PIXIS100, Teledyne Princeton Instruments).Samples were prepared in quartz cuvettes by suspending QDs in cyclohexane.The samples were excited using a 532 nm laser (CPS532, Thorlabs).Emission was collected and focused using two silver-coated off-axis parabolic mirrors, and it was filtered through an 800 nm longpass dielectric filter into a monochromator before imaging on the silicon camera.Photoluminescence quantum yield (PLQY) data were measured using a silicon photodiode (818-UV, Newport) coupled to a lock-in amplifier (SR830, Stanford Research), using chopped 405 nm laser excitation (LDM405, Thorlabs) and an optical integration sphere (RTC-060-SF, Labsphere). [39]icroneedle Fabrication: In this report, a 0.8 cm 2 , 10 × 10 microneedle array patch was used.The individual needles were square-based pyramid shapes with a height of 800 μm, base of 200 μm, and pitch of 500 μm.Commercially available negative PDMS microneedle molds were purchased from Micropoint Technologies.To load the fluorescent microparticles, 75 μL of an aqueous dispersion of QD-PMMA, QD-PLGA1, or QD-PLGA2 microparticles (5 mg mL -1 ) was added to the top of the negative microneedle mold and centrifuged at 3000 rpm for 5 min to concentrate the fluorescent microparticles at the needle tips.To ensure an even loading, the microneedle mold was then rotated 180°and centrifuged at 3000 rpm for another 5 min.To form the body of the microneedle, 100 μL of a 16 wt% PVA/sucrose (50:50 PVA:sucrose) solution was added to the microneedle mold and centrifuged at 4000 rpm for 10 min.PVA/sucrose was a proven matrix for the preparation of dissolvable microneedle arrays. [19,40,41]An additional volume of PVA/sucrose (100-300 μL) was added to the negative microneedle mold following centrifugation to minimize volume loss upon drying and to form the backing.After drying at room temperature for 24 h, an acrylic disk was attached to the back of the microneedle array using a double-sided tape.The acrylic disk allowed for easy handling and provided a large, flat surface to help with the microneedle application.The microneedle arrays were then removed from the mold and dried under high vacuum for a further 24-48 h.The fully formed microneedles now contained QD-PMMA microparticles concentrated at the needle tip and were ready for application.Following this procedure, the maximum possible loading of QD-PMMA per microneedle array was 0.375 mg which contained 0.225 mg of QDs as calculated from the wt % of the QDs in PMMA-QD particles (60 wt%).It was likely that the actual loading mass of the QD-PMMA particles was lower than this due to loss upon centrifugation and incomplete microneedle dissolution upon application.Once microneedle array patches were fabricated, they were stored and transported in designated packages to protect the microneedle tips and avoid damage.
Encoded Microneedle Fabrication: Encoded microneedle arrays were prepared in one of two ways.Either a scalpel was used to manually break individual microneedles from a fully formed array or sticky tape was used to block select microneedles in the negative molds prior to loading the fluorescent microparticles.
Hybrid Microneedle Fabrication: The select loading of QD-PMMA or QD-PLGA microparticles was achieved by blocking the desired microneedles in the negative molds with sticky tape prior to loading the fluorescent microparticles.First, QD-PMMA microparticles were loaded via centrifugation (as described above) to form the permanent sections of the hybrid designs (the secondary pattern).The sticky tape was then removed and QD-PLGA microparticles were loaded into the negative mold to form the complete, primary pattern.
Coulter Counter: To measure the size of the fluorescent microparticles, a small amount (≈5 mg) was resuspended in deionized water and measured using a Multisizer 3 (Beckman Coulter) with a 100 μm aperture.
SEM Microscopy: In preparation for SEM, samples were deposited on double-sided carbon tape and coated with a thin layer of Au/Pd using a Hummer 6.2 Sputtering System (Anatech) to prevent charging.Imaging was then performed using a JEOL JSM-5600LV scanning electron microscope with an acceleration voltage of 5 kV.
In Vitro QD-PLGA Microparticle Degradation Study: QD-PLGA microparticles (10 mg) were suspended in PBS (1 mL) and incubated at 37 °C on a rotating rack.At the desired time, aliquots were taken (≈ 100 μL) and analyzed via SEM to examine the extent of microparticle degradation.
[46][47][48] Separately, a QD-PMMA-loaded microneedle patch was applied to ex vivo human skin, and the QD signal brightness was quantified over 7 days.QD brightness was quantified using Image J (National Institutes of Health).
QD-PMMA and QD-PLGA Biocompatibility Study: HeLa cells were cultured in high glucose Dulbecco's modified Eagles medium with phenol red (DMEM, Invitrogen) supplemented with 10% FBS (Invitrogen) and 1% antibiotic (Invitrogen).Cells were seeded at a density of 5000 per well in a 96-well plate in a full-growth medium.Twenty hours after seeding, the media was replaced with fresh media including QD-PMMA and QD-PLGA microparticles at different concentrations, and the cells were incubated for 20 h.Then the media was removed, cells were washed once with PBS buffer, and 10% of the cells were used for Cell Counting Kit-8 (CCK-8) (Sigma-Aldrich, St. Louis, MO).Cells were incubated for 4 h and absorbance was measured at 450 nm.
Mechanical Compression Test: A compression force test was performed on QD-PMMA microparticle-loaded microneedle array patches to assess their mechanical robustness for proper skin penetration. [49]The pressure required to puncture human skin was known to be roughly 100 psi, which was equivalent to 0.689 MPa. [50,51]The minimum force required to puncture human skin with the microneedle patch (100 microneedles with a needle base dimension of 200 × 200 μm) was 2.756 N (Equation ( 1)), which was the compression force one microneedle array patch must endure to pierce the human skin.10 × 10 QD-loaded microneedle patches (n = 5) were compressed at a rate of 5 mm min -1 using Instron 5943 (Norwood, MA) (Movie S1, Figure S4, Supporting Information).The maximum load, load at break, and Young's modulus were measured with Instron static load cell (± 500 N) and Instron Bluehill 3 software.For all patches, the compression force measurements reached the upper limit of the load cell (500 N) before the platens reached the base of the needles, indicating that the microneedle patch could endure more than 500 N, easily exceeding the minimum force to endure human skin penetration.
As for the individual microneedles, the minimum force a needle must endure for human skin penetration was 0.027 N (Equation ( 2)).When the failure force test was repeated on individual microneedles (n = 4), the average maximum load experienced by a microneedle before breakage was 22.55 (±4.23)N (Figure S4, Supporting Information), which easily exceeded the minimum force required to endure.
Rat skin was much more fragile and easier to pierce compared to human skin, considering the average thicknesses of stratum corneum, viscoelastic epidermis, and dermis were 9.38, 23.58, and 382.42 μm for rats and 17.07, 99.80, and 2284.05μm for human, respectively.The Young's modulus of human skin was also higher than that of rats by 2-4 folds. [52]x Vivo Microneedle Application: Microneedle application and QD-PMMA microparticle delivery were examined using human cadaveric skin donated to the National Disease Research Interchange under protocol DLAR9-001.Microneedle application was performed using an MPatch spring-loaded applicator (Micropoint Technologies) that featured a spring speed of 1-2 m s -1 and an average spring impact force of 1.6 N. Once applied, the microneedle array was held in the skin for 2 min.The tissue was imaged before and after application using the NIR handheld imaging device (Movie S2, Supporting Information).
Microneedle Application to Wistar Rats: Microneedles containing fluorescent microparticles were applied to Wistar rats (Charles River Laboratories) which were between 8-12 weeks of age and weighed ≈250 g.Prior to application, the rats were anesthetized via continuous inhalation of isoflurane (2.5%).Once anesthetized, the hair on the rear flank of the Wistar rats was shaved with an electric razor and then treated with depilatory cream for 2 min.The site of application was then rinsed with water and sterilized with an ethanol swab.The microneedles were applied for 2 min using the spring-loaded applicator.
Imaging with Handheld Device: Imaging with the handheld device was performed on rats that were anesthetized via continuous inhalation of isoflurane (2.5%).To minimize user error, the imaging wand was clamped in a retort stand.The rat was moved to the center of the pattern on the rat skin with the light beam (significant variation in light intensity was observed when the pattern was not placed in the center of the light beam).At each imaging session, images were taken at different gains, exposure times, and contrast to capture a range of images and monitor the pattern over time.
Pattern Recognition/Machine Learning Algorithm: The image classifier based on a convolutional neural network was developed using a transferbased learning approach, in which only the last three layers of a pre-trained AlexNet convolutional neural network were modified to incorporate the new patterns.This emerging approach helps avoid huge computational costs involved in re-training the AlexNet which was originally developed with 60 million parameters and 650 000 neurons. [38]Briefly, a maximum epoch number of 20, 5250 iterations per epoch, and a learning rate of 0.0001 were used.The first 22 layers of a pre-trained AlexNet CNN were kept intact, while only the last three layers were used to train to a specific set of patterns.Two different neural networks were trained based on the patterns used for PMMA or PLGA particles.
In this study, a sophisticated algorithm was developed for the generation of augmented synthetic patterns for the training.The algorithm was able to make synthetic microneedle patches with an arbitrary n × m array of dots, representing the needles.The light intensity of each individual dot followed a Gaussian distribution which meant that the highest intensity was allocated to the center of dots, while the neighboring intensity dropped at a Gaussian rate moving away from the center.The radius within which the intensity dropped was specified randomly per image.To generate synthetic patterns covering all potential case scenarios in real on-field imaging, a wide range of random modifications was performed to the background and the pattern.The purpose of these modifications was to enhance detection reliability and robustness by minimizing any artificial bias in classification due to the noise or location-based sensitivity.Briefly, random noise was applied to the background, as well as random Gaussian wrapping of the patterns, followed by scaling the image intensity.More detailed considerations were further made to make sure the resulting wrapped patterns would always stay within the image frame.The resulting images were augmented to capture different imaging angles, different gains, and the presence of noise in the background.Before the training, the resulting images were subsequently turned to a gray-scale at the size of 227 × 227 pixels suitable for AlexNet.
Two types of synthetic images, namely, perfect and imperfect patterns were generated (Figure S12, Supporting Information).The former corresponded to images where all the dots (i.e. needles) were perfectly present in the patch.The imperfect patterns referred to the case where a random number of dots was missed during the transfer.The aim of imperfect patterns was to address potential misclassification caused by missed needle transfer during real applications in vivo.As such, a random 10% probability was specified for each dot-independently defined-for being missed during the transfer, in the array.Both types of synthetic images were incorporated into the network to further improve the reliability and robustness of classifications.Accordingly, each training was composed of 10K perfect images combined with 15K imperfect ones (Figure S13, Supporting Information).As such, 70% of images were used for training and 30% for validation.The validation accuracy for PMMA, and PLGA particles after the 20 th epoch was found as 99.42%, and 99.71% respectively.Testing was performed directly on in vivo images (image format of JPG) each having a single pattern per image.Before performing classifications, to further im-prove the accuracy, an imaging algorithm was employed to automatically crop around the pattern within the background.The cropped image was subsequently turned into a gray-scale, resized to the default input size for AlexNet (227 by 227), and its intensity was normalized.Further explanation of the preparations done on the test images can be found in Figure S12, Supporting Information.All image processing and machine learning codes were developed in MATLAB using AlexNet Pretrained Toolbox and ran on a single GPU (Nvidia GeForce GTX1080).
The resulting classifiers for each set of PMMA or PLGA patterns represented excellent detection accuracy within the intended time (Figure 4).All images were detected correctly with at least 98% probability of correct identification.The high accuracy was notable regarding the complexity of the patterns and random variations in the imaging angles as seen in the in vivo images.Further, it was demonstrated that the synthetic data generator was successfully able to produce reliable images to feed the training and validation sets with high accuracy.More importantly, it was demonstrated that these classifiers could be trained without the need to physically make the patterns.This approach provided a future roadmap toward design optimization of the coding scheme, without the need to fabricate the patches, to further improve the long-term reliability of the encoded data.

Calculation of the Number of Unique Patterns for Information Encoding:
The number of unique patterns that could be generated using the microneedle array was calculated according to the formula for combinations without repetition (Equation ( 3)).
where n is the number of elements to choose from (15) and k is the number of elements chosen (1, 2, 3, or 4).The total number of patterns available (32767) was the sum of the number of patterns when k was equal to 1, 2, 3, or 4. Statistical Analysis: Statistics analyses were performed using Graph-Pad Prism software using a two-tailed Student's t-test for pairwise comparisons (non-statistical significance p > 0.05).

Figure 1 .
Figure1.A) Dissolvable microneedle arrays can transfer a specific pattern of invisible-to-the-naked-eye fluorescent microparticles to the skin of the patient.The pattern can be read with a handheld imaging device to retrieve information.B) The microneedle array can be modified to achieve a change in the pattern after a certain amount of time by the selective loading of biodegradable fluorescent microparticles (as indicated by the green dot).This is used to encode information in both a spatial and temporal manner.

Figure 2 .
Figure 2. A) CuInSe2-core and ZnS:Al-shell QD nanoparticles with oleic acid surface functionalization are encapsulated in biocompatible polymer PMMA to form QD-PMMA microparticles.B) Photoluminescence spectra of the QD and the QD-PMMA microparticles.Encapsulation shows little effect on the photoluminescent properties of the QDs.C) Image of microneedle array, scale bar = 1 cm.D) Image of microneedle array taken with handheld NIR imaging device.E) SEM of QD-PMMA microparticles, scale bar = 3 μm.F) Section of microneedle array showing the QD-PMMA microparticles localized in the needle tips, scale bar = 0.5 mm.G) Encoding scheme.Fifteen 2 × 2 squares of microneedles are available for encoding.The top right square is removed for pattern alignment and the outer edge of the arrays is kept permanent.H) Longitudinal imaging of three QD-PMMA patterns (PMMA1-3) in Wistar rats over 180 days.

Figure 3 .
Figure 3. A) Photoluminescence spectrum of unencapsulated QDs (blue) and QDs encapsulated in PMMA (red) and PLGA (green).B) SEM imaging of QD-PLGA microparticles incubated in PBS at 37 °C at days 0, 16, and 41, scale bar 10 μm.C) Longitudinal imaging of a representative QD-PLGA pattern in vivo using the handheld imaging device.D) Two hybrid, shape-changing pattern designs and in vivo imaging showing the distinct change in pattern after 3 months.

Figure 4 .
Figure 4. A) Longitudinal imaging of four PLGA patterns over 118 days, demonstrating the temporary lifetime of the encoded information when the QDs are encapsulated within a biodegradable polymer.B) Probability of accurate pattern identification using the custom neural network.All patterns could be accurately identified with a very high probability for the lifetime of the pattern until the PLGA patterns degraded.