For transdermal patches, the release of active ingredients from the matrix is a prerequisite for the formulation to exert its pharmacological effects. This process is called release for topical preparations or dissolution for solid preparations [1]. The in vitro release rate is the key quality attribute of transdermal patches, and is currently the most important patch quality evaluation index [2]. The traditional in vitro release test (IVRT) of patches is usually carried out with the help of experimental devices such as paddle dish method, rotating cylinder method, reciprocating scaffold method, etc. These methods carry out in vitro release experiments in specific release medium according to the specified duration, and determine the drug release fraction in the release medium with the help of high-performance liquid chromatography (HPLC) or ultraviolet–visible spectroscopy (UV–Vis). Finally, use the drug release model equation to directly plot or fit the in vitro drug release curve. At present, the most widely used method for determining the in vitro release of transdermal patches is the paddle dish method specified in the United States Pharmacopeia (USP), but there are many inconveniences in the actual operation of this method. For example: (1) the paddle dish method requires a large amount of release medium to be configured, and requires degassing and maintaining a constant temperature, making the operation process more complicated, and frequent sampling can interfere with the stability of the release environment and affect experimental results. (2) In addition, due to the method of use, the single test cycle of in vitro release of patches is long. For example, the single test cycle of in vitro release of estradiol sustained-release patches included in the Chinese Pharmacopoeia (ChP) 2020 edition even reaches one week, which takes a very long time. In order to overcome these shortcomings, many researchers are focusing on the related research of using spectral analysis technology that does not require complex sample pretreatment to predict the drug release profile in vitro [3], [4], [5], [6]. The advantage of spectral analysis technology is that it can quickly and non destructively obtain the physical and chemical information of various components on the surface and inside of the preparation, such as near infrared (NIR) or Raman spectra. After using spectral mapping technology to collect sample information from multiple sites on the surface of the samples, combined with multivariate analysis methods, such as Multivariate Curve Resolution (MCR), partial least squares regression (PLSR) and artificial neural network (ANN), the information contained in the spectra can be fully used to predict the dissolution behavior of the preparation. Therefore, spectral mapping technology combined with multivariate data analysis is a feasible and potential new method. In the previous study of our research group [7], the in vitro dissolution behavior of sinomenine hydrochloride tablets was successfully predicted based on Raman mapping technology combined with multivariate analysis method. Although there are few such studies on patches at present, in fact, spectral mapping technology is also suitable for the field of transdermal patches [8], [9], [10], [11].

As a rapid, non-destructive analysis technology that does not require complex pretreatment of test samples, Raman mapping technology has great advantages in the analysis of drug preparations [12], and it allows researchers to probe the microscopic morphology of the drug surface when combined with microscopy, which not only significantly improves the spectral intensity, but also enables the quantification of lower content of the drug substance and the characterization of the drug distribution on the surface of the drug product. Therefore, Raman mapping technology has been used in the analysis of a variety of drug dosage forms [13], [14], [15], [16], [17], [18], [19], [20], in view of the shortcomings of the IVRT methods for transdermal patches, and combined with the advantages of Raman mapping technology which is fast, nondestructive and does not require complex sample pretreatment, this study used Raman mapping technology combined with multivariate analysis method to predict the in vitro release behavior of diclofenac sodium transdermal patches. In fact, many researchers have applied Raman mapping technology to the quality control of topical preparations. For example, some researchers used Raman spectral microscope to locate the active pharmaceutical ingredients (API) in transdermal patches [8]; Some people used confocal Raman microscope to identify the state of different kinds of API in the matrix. The experimental results show that Raman spectra are suitable for identifying API in the preparation [21]; In addition to quantifying API in transdermal patches and detecting the state of API in the matrix, the detection of API crystals in patches is also an important research direction [22], [23], [24], the experimental results showed that Raman spectra might detect subtle changes in crystal morphology; Some researchers used confocal Raman spectra combined with chemometric methods such as MCR to image the crystal distribution and morphology of selected areas on the surface of different brands of commercially available fentanyl transdermal patches [18], and more and more researchers have used confocal Raman microscopy to directly detect the penetration of drugs in human skin, which is of great significance for the quality evaluation and development of transdermal patches [25]. To sum up, there have been many related studies on the application of Raman spectra in the quality control of transdermal patches.

Therefore, in this study, Raman mapping technology was used to collect the spectrum of the surface of the commercially available diclofenac sodium patches, and then the paddle dish method was used to determine its in vitro release at a specific time point as a standard value. Finally, PLSR and convolutional neural network (CNN) algorithms were used to establish models to predict the in vitro release of patches. The main contents are as follows: 4 batches of diclofenac sodium patches commercial finished products of different production batches were purchased. 25 samples were taken from each batch to form a sample group with a total number of 100. Raman spectra of a fixed area on the surface of each sample were collected using a confocal Raman microscope in multipoint mode. After spectral data were obtained, the in vitro release of all samples were measured using the paddle dish method, and the in vitro release curves were fitted. After processing the Raman spectral data, the in vitro release prediction models of patches were established using curve fitting dependent (CFD) and curve fitting independent (CFI) methods [26], and each method established three models. Finally, the performance of the built models was evaluated using the model prediction accuracy and other indicators, and the advantages and disadvantages of different prediction methods were compared, providing a new idea for the establishment of patch quality evaluation methods.