Introduction

Hyperspectral imaging (HSI) is an optical method that simultaneously acquires both spatial and spectral information across 30 to 100 spectral channels1. Compared to conventional color imaging, HSI has been known to significantly enhance contrast and differentiation of the target based on its spectral characteristics2. Recent advances in optical systems, big data acquisition, and deep learning-based hyperspectral image analysis have led to the practical application of HSI across various fields, including remote sensing3, medicine4,5, and industrial inspection6,7. To acquire hyperspectral images, three methods are typically employed: spatial scanning, spectral scanning, and snapshot imaging1,8. Each method involves trade-offs among spatial resolution, spectral resolution, and acquisition speed, necessitating the selection of an appropriate technique depending on the imaging purpose and environment. However, in conventional HSI systems, the spatial and spectral resolution as well as acquisition speed are largely constrained by the optical setup. Therefore, in dynamically changing environments or when the target’s shape or position varies over time, conventional HSI systems face limitations in efficient hyperspectral data acquisition.

In this study, we present a novel hyperspectral imaging system employing a digital micromirror device (DMD) to achieve high spectral resolution with variable spectral data acquisition regions. The DMD consists of an array of micro-scale mirrors, each of which tilts between two discrete angles depending on the input signal, functioning in a binary mode. DMDs are widely used in beam projectors9 and other optical applications such as lithography10,11 and digital holography12,13,14,15, due to their ability to dynamically control light reflection patterns without mechanical movement of optical components. We exploited the DMD to measure spectral data at dynamically adjustable regions with high spectral resolution by overcoming the limitations of conventional spatial scanning methods.

In spatial scanning methods, acquisition of hyperspectral images with high spectral resolution requires the use of dispersive optical components such as prisms or gratings16,17. Traditional spatial scanning HSI systems rely on such components to obtain spectral data at a single point or line, followed by mechanical scanning to cover a wide field of view. While this enables high spectral resolution, it becomes problematic when the target is moving or when it is difficult to precisely locate the measurement positions, leading to challenges in accurate image reconstruction18,19. Previous studies also utilized the DMD by harnessing a single or lines of DMD pixels as virtual scanning unit without physically moving the optics20,21,22,23. Nonetheless, studies using DMDs to acquire high spectral resolution HSI in dynamic environments remain limited.

To implement a new hyperspectral imaging method that shows high spectral resolution with dynamically adjustable spectral data acquisition regions, we propose a system in which the DMD selectively reflects light from a target region toward a spectrometer for hyperspectral measurement, while reflecting light from the remaining pixels directly toward a CMOS camera to simultaneously capture a wide-field image. The proposed method enables flexible selection of the spectral measurement area, while also providing spatial information of spectral data acquisition regions without the need for additional image reconstruction processes. Moreover, by utilizing the spectrometer, the proposed system shows the advantage of high spectral resolution. The proposed method could be a versatile tool in applications that require high spectral resolution hyperspectral imaging under dynamically changing conditions.

Results

Concept of the DMD-based hyperspectral imaging system

Figure 1(A) illustrates the schematic concept of the proposed method. The DMD operates in a binary mode, where each micromirror can tilt between two discrete angles based on the input signal. By exploiting this property, spectral data and wide-field image can be acquired simultaneously: light reflected from the “on” state pixels is directed to a spectrometer for spectral measurement, while light from the “off” state pixels is captured by a color camera for wide-field imaging. As the DMD pixels assigned to spectral and imaging measurements are mutually exclusive, the spectral acquisition region appears as a distinct dark area within the wide-field image.

Fig. 1
Fig. 1
Full size image

Schematics of DMD-based hyperspectral imaging system with selective spectral imaging area. (A) Spectral image was acquired from DMD pixels in the “On” status, while the corresponding wide-field image is simultaneously captured from the “Off” state pixels, allowing the verification of the spectral acquisition region. (B) A representative demonstrating selective spectral imaging using the proposed DMD-based approach, in which the acquisition area is clearly visualized as a dark region in the wide-field image.

To exploit the DMD as a selective spectral imaging system, the sample image is first projected onto the DMD surface. A binary pattern corresponding to the region of interest (ROI) is then uploaded to the DMD, enabling localized spectral data acquisition alongside wide-field imaging, as shown in Fig. 1B. The spectral acquisition area can be arbitrarily defined by modifying the binary pattern, and its location is clearly visible in the wide-field image as a dark region, confirming the selective spectral measurement capability of the system.

Technical evaluation of the proposed method

Figure 2A presents the optical setup of the DMD-based hyperspectral imaging system. Detailed specifications of the optical components are provided in the Methods section. In brief, the sample image is projected onto the DMD using a 4f optical system consisting of two lenses. Depending on the uploaded binary pattern, DMD pixels in the ‘on’ state direct reflected light to a spectrometer for spectral data acquisition, while pixels in the ‘off’ state reflect light to a CMOS camera for wide-field imaging. As the DMD functions similarly to a 2D diffraction grating, broadband illumination results in complex diffraction artifacts24,25. To minimize these effects, all optical paths were constructed using 4f configurations to ensure consistent imaging performance.

Fig. 2
Fig. 2
Full size image

Technical evaluation of the DMD-based hyperspectral imaging system. (A) Schematic of the optical setup of the DMD-based hyperspectral imaging system. L; Lens, M; Mirror, AOTF; Acousto-optic tunable filter, DMD; Digital micromirror device. (B) Spectral profiles of the input light source measured before and after reflection from the DMD, showing consistency in spectral shape with a slight reduction in intensity due to reflection losses. (C) Spectral data acquired by the spectrometer under four representative DMD patterns. Left: corresponding wide-field images; Right: spectral profiles measured at each selected region. Scale bar is 0.5 mm. (D) Representative wide-field images captured at five central wavelengths (450, 500, 550, 625, and 700 nm) with a 5 nm bandwidth. The overlapped image was generated by averaging the four spectral images to assess wavelength-dependent imaging consistency. Scale bar is 0.5 mm.

The key concept of the proposed system is to simultaneously acquire spectral information alongside the corresponding spatial information in a wide-field image, effectively indicating the precise location of spectral measurements. Therefore, stability and consistency of both spectral and wide-field imaging performance are essential. To validate the spectral fidelity of the system, the spectrum of the supercontinuum light source was measured both before and after reflection from the DMD. As shown in Fig. 2B, the light source exhibits a ~ 250 nm bandwidth with peak intensity around 600 nm. This spectral bandwidth was primarily determined corresponding to operational wavelength limitations of both the DMD and the AOTF. While power loss was observed after DMD reflection, the overall spectral shape and bandwidth remained consistent, confirming minimal distortion.

Figure 2C shows four representative DMD patterns (left) and their corresponding measured spectra (right). Although the intensity levels vary depending on the pattern, the spectral profiles (in terms of shape and bandwidth) remain consistent across patterns, indicating no distortion or wavelength-dependent bias in spectral acquisition. The observed intensity variation is due to the Gaussian distribution of the input laser power.

Next, we examined the impact of broad bandwidth of light signals on wide-field imaging performance using the supercontinuum laser with a 5 nm bandwidth. Imperfect optical alignment could introduce wavelength-dependent variations in wide-field imaging, potentially degrading spatial resolution when a broadband source is used. To assess this, the same binary pattern was uploaded to the DMD, and wide-field images were acquired at 450, 500, 550, 625, and 700 nm with a spectral bandwidth of 5 nm. Figure 2D shows these representative images, and the overlapped image demonstrates excellent alignment and image clarity across wavelengths, confirming robust wide-field imaging across the visible spectrum.

Assessment of the accuracy of spectral data acquisition

To qualify as a novel hyperspectral imaging technique, the proposed method should be capable of acquiring spatially resolved spectral information. To validate this capability, a 1951 USAF resolution target was used due to its well-defined spatial features. The vertical line patterns corresponding to Group − 1, Element 6 were overlaid with red, green, and blue cellulose paper, respectively, to introduce spatially distinct spectral characteristics (Fig. 3A). Spectral information from each of these color regions was acquired in conjunction with their corresponding wide-field images.

Fig. 3
Fig. 3
Full size image

Evaluation of spatially resolved spectral imaging capability. (A) A color sample was prepared by covering red, green, and blue cellulose papers to the 1951 USAF resolution target. (B) DMD patterns (top) and corresponding wide-field images (bottom) measured by the CMOS camera. Scale bar is 2 mm. (C) Spectral profiles of red, green, and blue cellulose papers measured by using DMD patterns shown in (B). (D) Spectral mixing analysis using two DMD patterns. The black line shows the measured spectrum from a mixed DMD pattern, and the yellow line shows the numerical summation of individual spectra from red and blue regions. (E) Spectral profiles measured by varying the mixing ratio (25%, 50%, 75%, and 100%) of two DMD patterns.

Figure 3B shows the DMD binary patterns and the corresponding wide-field images. Pattern 1 directed all light toward the color CMOS camera to obtain a full wide-field image, while Patterns 2–4 selectively reflected light from the red, green, and blue regions, respectively, to the spectrometer for spectral data acquisition. The measured spectral profiles for each region are shown in Fig. 3C, confirming that the proposed method successfully captured spatially resolved spectral signatures from the designated ROI.

To further demonstrate the method’s ability to acquire spectral information from arbitrarily selected spatial areas, we tested the system by combining two DMD patterns and compared the resulting spectrum with the linear combination of the individual spectra obtained from those regions. As shown in Fig. 3D, the spectral profile acquired by using the mixed DMD pattern (black line) closely matched the summed profile of the individually measured spectra (yellow line), indicating accurate additive spectral behavior and spatial selectivity.

Spectral unmixing is a key process in hyperspectral imaging, wherein a composite spectral signal is decomposed into constituent endmembers with corresponding proportions. This necessitates that the acquired spectrum should vary proportionally with the spatial composition of selected regions. To verify this, we gradually changed the ratio of Pattern 2 (red) and Pattern 4 (blue) and measured the resulting spectra. The spectral profiles shown in Fig. 3E demonstrate a smooth and consistent transition, validating that the proposed system can reliably perform spatially resolved spectral acquisition and is suitable for spectral unmixing applications.

Clinical application of the proposed method

To evaluate the biomedical applicability of the proposed method, a histological tissue slide sample of gastric cancer was analyzed. Figure 4A shows an H&E-stained image of the sample, with blue and red outlines indicating regions of normal and cancerous tissue, respectively, as identified by a pathologist. For validation, the sample was first imaged using a liquid crystal tunable filter (LCTF)-based hyperspectral microscope employing the spectral scanning method. The averaged spectral profiles of the normal and cancer regions obtained via the LCTF-based hyperspectral microscope are shown in Fig. 4B.

Fig. 4
Fig. 4
Full size image

Clinical application of the proposed selective spectral imaging method. (A) RGB image of an H&E-stained tissue slide. The white dashed box indicates the wide-field imaging area; red and blue lines outline cancerous and normal tissue regions, respectively. (B) Ground-truth spectral profiles of cancerous and normal tissues. (C) Measured wide-field images at two DMD patterns targeting cancerous and normal tissue areas. Scale bar is 0.5 mm. (D) Spectral profiles of cancerous and normal tissues acquired using the proposed method.

Subsequently, the same tissue slide was measured using the proposed DMD-based hyperspectral imaging system. Figure 4C shows the wide-field image acquired using this system, where the DMD patterns were selectively applied to the ROI based on the histopathological annotation. The corresponding spectral profiles of the normal and cancer areas, as measured by the proposed method, are shown in Fig. 4D. Although the absorbance values of normal and cancer tissues in Fig. 4B and D show slight differences, the overall trend which is higher absorbance in cancer tissues remains consistent. These discrepancies in absolute absorbance values may arise from differences in imaging conditions such as magnification, resolution, and hyperspectral imaging methodology. Consequently, many studies emphasize the use of normalized spectral signatures, or the shape of the spectral profile, over absolute intensity values to ensure robust and reliable analysis across varying acquisition conditions26,27. Therefore, the consistent measurement of spectral signatures in this study validates both the spectral accuracy and the practical applicability of the proposed approach.

These results demonstrate that the proposed DMD-based hyperspectral imaging method provides comparable performance to conventional hyperspectral microscopy techniques and holds promise for biomedical applications, including disease diagnosis based on intrinsic tissue optical properties.

Discussion

In this study, we developed a novel hyperspectral imaging technique that enables flexible selection of spectral acquisition regions using a DMD while simultaneously acquiring wide-field images to identify the corresponding measurement locations. The proposed method enhances the advantages and mitigates the limitations of conventional spatial scanning hyperspectral imaging methods. The specific benefits of this method include: (1) achieving spectral resolution comparable to that of spatial scanning hyperspectral imaging techniques, (2) precisely identifying spectral acquisition regions without the need for post image registration process, using the darkened regions in the simultaneously captured wide-field images, (3) flexibility in selecting ROIs via customizable DMD patterns, allowing for free adjustment of both the shape and size of the spectral acquisition areas. Unlike conventional methods that require post image processing to select target regions after capturing the entire hyperspectral image, the proposed method enables real-time spectral measurement only from designated regions, significantly accelerating the analysis process. (4) capability to acquire RGB images by directing all DMD pixels to the CMOS camera, with on-demand spectral measurements performed by simply uploading appropriate DMD patterns to the target regions. This allows hyperspectral and RGB imaging to be achieved using a single compact system, without the need for an additional complex optical setup.

While the proposed method presents multiple advantages and promises to be a new approach in hyperspectral imaging, several challenges remain to be addressed. First, since the DMD consists of an array of micromirrors, it functions as a 2D grating, resulting in complex diffraction phenomena—particularly across a broad spectral range. If the pitch size of the DMD decreases, then diffraction effects become more pronounced. These effects become especially problematic in the Fourier plane, making optical alignment difficult. To overcome this, we implemented a 4f optical system between the DMD, the spectrometer, and other components to ensure image-plane-to-image-plane transfer and suppress high-frequency diffraction artifacts. However, this design also results in optical power loss, as high-frequency information is filtered out by optical elements such as lenses that act as finite apertures that block these higher-order diffraction components. In addition, an integrating sphere was employed in front of the spectrometer to ensure uniform and stable spectral acquisition. Although this approach is effective, the integrating sphere introduces additional power loss as the incoming light is distributed across the internal surface area of the sphere. This limitation could potentially be addressed by replacing the integrating sphere with a multimode fiber to improve light collection efficiency. Beyond optical power loss, off-axis optical configurations may introduce field curvature, which can degrade image quality across the field of view. This optical aberration poses a challenge, particularly in wide-field imaging applications. However, recent advances in multi-color holographic projection techniques offer a promising solution to mitigate this issue by correcting field curvature through computational or optical wavefront control28. Integrating such approaches into the current system may enhance overall image fidelity and uniformity, further expanding its applicability to high-resolution hyperspectral imaging.

Furthermore, to obtain wide-area hyperspectral data, the system would still require raster scanning or line scanning across all DMD pixels, similar to traditional spatial scanning techniques. In the current study, all experiments were performed by projecting a single static pattern onto the DMD during each measurement, primarily to demonstrate the proof-of-concept feasibility of the approach. As such, dynamic or real-time environments were not evaluated. To address this limitation, we propose the implementation of a dynamic scanning strategy. Initially, a large scanning unit can be used to broadly acquire hyperspectral data, and if target spectral features are detected, a smaller scanning unit can be applied to those regions for high-resolution measurement. As DMD pixel switching can operate over several kHz speeds25, this dynamic scanning approach could offer significantly faster imaging compared to conventional spatial scanning systems.

Although some challenges remain, the proposed method offers a promising solution to overcome the fixed spatial and spectral resolution limitations of current hyperspectral imaging modalities (spatial scanning, spectral scanning, and snapshot methods). By enabling simultaneous acquisition of spectral and wide-field images, it provides accurate localization of spectral measurements, a critical feature for real-time imaging applications. For example, in endoscopic applications, where clinicians must constantly adjust their targets, the proposed method’s flexibility in selecting and resizing measurement regions makes it especially advantageous. Furthermore, the system is not inherently limited to the visible wavelength range. Although the present experiments were conducted within the visible spectrum due to the wavelength-specific design of the light source, the DMD, and optical components, the method is readily extendable to other spectral regions such as the near-infrared or infrared by exploiting compatible optical elements. Given its dynamic spectral imaging capabilities enabled by the DMD, the proposed hyperspectral imaging method holds great potential for diverse applications in medical imaging, industrial inspection, and remote sensing as a next-generation hyperspectral imaging approach.

Methods

Hyperspectral imaging system

The DMD-based hyperspectral imaging system consists of a DMD (Texas Instrument, DLPLCR6500), a spectrometer (Avantes, AvaSpec-ULS4096CL-EVO), and a CMOS camera (FLIR, GS3-U3-51S5C-C), a supercontinuum laser (NKT Photonics, SuperK FIANIUM FIU-6), equipped with an acousto-optic tunable filter (NKT Photonics, SuperK VARIA), was employed as the light source. This source allows spectral tuning of the central wavelength from 420 to 720 nm with a variable spectral bandwidth ranging from 1 nm to 250 nm, enabling validation of the system’s spectral imaging capabilities.

The sample was imaged using two plano-convex lenes with focal lengths of 75 and 180 mm to project onto the DMD panel. Depending on the tilt angles of the DMD pixels in the “On” and “Off” states, light was directed toward either the spectral or wide-field imaging paths. For spectral imaging, two plano-convex lenses with focal lengths of 100 mm and 75 mm were used to demagnify and deliver the reflected light from the DMD panel into an integrating sphere (Thorlabs, IS200), which was connected to the spectrometer. For wide-field imaging, a plano-convex lens with a focal length of 125 mm and an imaging lens with a focal length of 50 mm were positioned in front of the CMOS camera.

Image acquisition and processing processes

The patterns uploaded to the DMD were generated using MATLAB 2024R to define the regions for spectral data acquisition. Once a pattern was uploaded, spectral and wide-field images were simultaneously acquired with and without the sample, enabling the measurement of both the sample’s spectral response and the light source reference. In addition, dark spectral signals were collected under light-blocked conditions to account for background noise.

Reflectance and absorbance were calculated using the following equations:

$$\:Reflectance\left(\lambda\:\right)=\:\frac{Sample\left(\lambda\:\right)-Dark\left(\lambda\:\right)}{White\left(\lambda\:\right)-Dark\left(\lambda\:\right)}$$
$$\:Absorbance\left(\lambda\:\right)=\:-{\text{log}}_{10}\left(Reflectance\:\left(\lambda\:\right)\right)$$

where sample(λ), white(λ), and dark (λ) were spectral signals obtained from a sample, white reference target (Spectralon, Diffuse Reflectance Standard), and dark condition at the wavelength of λ.

1951 USAF resolution target sample

For the technical evaluation of the proposed method, a 1951 USAF resolution target (Thorlabs, R3L3S1N) was used. The three vertical lines corresponding to Group − 1, Element 6 were covered with cellulose papers in red, green, and blue to generate spatially distinct spectral regions. Although the resulting wide-field images of the cellulose papers appeared slightly unclear due to the uneven surface texture of the cellulose paper, they were sufficient to evaluate the spectral imaging capability of the proposed method.

Tissue slide sample

The tissue slide used in this work was randomly selected from surgically resected specimens of a patient diagnosed with advanced gastric cancer at Ajou University Hospital as part of a routine clinical pathology investigation. The sample was anonymized prior to use. The use of tissue samples in this study was approved by the Institutional Review Board of Ajou University Hospital (IRB approval no. AJOUIRB-SMP-2021-172). Informed consent was obtained from the patient and all methods were carried out in accordance with relevant guidelines and regulations. The tissue slide from gastric cancer was stained with hematoxylin and eosin (H&E) using standard histological protocols. Formalin-fixed, paraffin-embedded Sect. (4 μm thick) were deparaffinized in xylene, rehydrated through graded alcohols, stained with hematoxylin, differentiated in acid alcohol, and blued in running tap water. Following eosin staining, the slide was dehydrated, cleared in xylene, and mounted with coverslip using a synthetic mounting medium.