DART: Handheld DUV Ptychographic Imaging

• Deep-ultraviolet Ptychographic Pocket-scope (DART) is a handheld imaging device enabling lensless, label-free molecular mapping via intrinsic DUV spectroscopic contrast.
• It employs multi-mode ptychography and an innovative virtual error-bin method to achieve sub-diffraction resolution and high-fidelity molecular mass maps.
• The system is applicable to cytopathology, neural tissue imaging, and blood cytometry, offering quantitative virtual staining and extensive fields of view.

The Deep-ultraviolet Ptychographic Pocket-scope (DART) is a handheld mesoscale imaging platform designed for lensless, label-free molecular mapping using intrinsic deep-ultraviolet (DUV) spectroscopic contrast. DART utilizes biomolecular absorption signatures at DUV wavelengths in combination with multi-mode ptychographic microscopy principles, achieving sub-diffraction resolution and molecular specificity across large fields of view and extended depth-of-field, all within a compact, portable form factor. The system’s algorithmic architecture incorporates a virtual error-bin methodology to mitigate systematic errors from limited temporal coherence and optical imperfections, producing high-fidelity molecular mass maps suitable for downstream virtual staining and quantitative analysis.

1. Optical and Mechanical Design

DART’s optical subsystem uses two DUV LEDs—one at 266 nm (Crystal IS KL265-50T-SM-WD) and one at 280 nm (Nichia NCSU334A)—to probe intrinsic biomolecular absorption. A 405 nm laser diode (D405-20, US-Lasers) is employed for high-coherence phase imaging. The beam path is folded with UV-enhanced aluminum mirrors defining a 20 cm optical path, ensuring uniform, refractive-optics-free illumination over centimeter-scale sample areas.

Samples are mounted 0.2–2 mm above a disorder-engineered random phase/amplitude coded surface (on fused silica), which is permanently bonded to a modified CMOS detector (Sony IMX 226)—its microlens array removed to double DUV sensitivity. The coded surface-sensor spacing is rigidly fixed at d₂ = 0.84 mm, leveraging the Fresnel regime for millimeter-range depth-of-field.

Sensor-shift ptychography is implemented by rastering the integrated coded sensor using voice-coil actuators and magnetic ball bearings in a randomized 1–3 μm grid. Two constraint rails prevent sensor rotation; sub-pixel motion is later recovered computationally. All optical, actuation, and sensor components are integrated in a compact, pocket-sized enclosure.

2. Measurement Model and Forward Operator

DART’s lensless ptychographic measurement is mathematically modeled by considering probe-sample interaction at each actuator position. For a coherent source, the intensity at scan position $R_j$ is

$I_j(q) = |F\{ P(r) \cdot O(r - R_j) \} |^2$

where $P(r)$ is the complex probe (illumination × coded surface), $O(r)$ is the sample’s complex transmission, $q$ is the sensor-plane coordinate, and $F$ denotes the 2D Fourier transform.

DUV LED illumination is spatially partially coherent, motivating an incoherent mode decomposition:

$$I_j(q) = \sum_s |F [ E_s(r - R_j) \cdot C(r) \cdot O(r) ] |^2$$

Here $E_s$ are the spatially incoherent modes, $C(r)$ is the coded surface modulation, and the sum runs over incoherent modes $s$.

3. Reconstruction Algorithm and Virtual Error-bin Mitigation

DART reconstruction utilizes a multi-mode ptychographic phase retrieval engine—specifically, an extended PIE/rPIE core—with a secondary virtual error-bin state that absorbs persistent model residuals arising from limited temporal coherence, multi-reflection fringes, and mask non-idealities. The observed intensity at pixel $(x, y)$ is modeled as

$$I_j(x,y) = \sum_s \left(|O_s(x,y)|^2 + |V_s(x,y)|^2\right)$$

with \begin{align*} O_s(x,y) &= [E_s(x - x_j, y - y_j) \cdot C(x,y)] \otimes PSF_{free}(d_2) \ V_s(x,y) &= [E_s(x - x_j, y - y_j) \cdot C_v(x,y)] \otimes PSF_{free}(d_2 \cdot a) \end{align*} where $a \approx 1.1$ slightly perturbs the propagation distance for the virtual state ($V$), and $PSF_{free}(d)$ is the Fresnel kernel. $C_v$ may be identical to $C$.

The multi-stage algorithm initializes $O$, $V$, and $E_s$ uniformly, and for each iteration and scan position, propagates and updates using PIE-style rules with amplitude correction and support or positivity constraints. A multiresolution approach refines the object at progressively higher spatial scales. Empirical reconstruction times are ~5 min for 400 DUV patterns (MATLAB, 4-core CPU) and ~40 s for the 405 nm single-mode case.

4. DUV Spectroscopic Molecular Contrast

DART exploits the characteristic DUV absorption of nucleic acids and proteins: nucleic acid absorption peaks near 260 nm, proteins near 280 nm. Reported decadic molar extinction coefficients (ε, in L·mol⁻¹·cm⁻¹) are

• $\epsilon_{nuc}(266) \approx 12,000$, $\epsilon_{pro}(266) \approx 8,000$,
• $\epsilon_{nuc}(280) \approx 6,000$, $\epsilon_{pro}(280) \approx 20,000$.

From measured sample and background intensities, the Beer–Lambert law yields pixel-wise optical density (OD):

$$OD_\lambda(x,y) = \log_{10}\left[\frac{I_{bg}(x,y)}{I_{sample,\lambda}(x,y)}\right]$$

A 2×2 system relates OD at both DUV wavelengths to local nucleic acid $(n_{nuc})$ and protein $(n_{pro})$ areal densities: \begin{align*} n_{pro}(x,y) &= \frac{OD_{266}\, \epsilon_{nuc}(280) - OD_{280}\, \epsilon_{nuc}(266)}{\epsilon_{pro}(266)\, \epsilon_{nuc}(280) - \epsilon_{pro}(280)\, \epsilon_{nuc}(266)} \ n_{nuc}(x,y) &= \frac{OD_{280}\, \epsilon_{pro}(266) - OD_{266}\, \epsilon_{pro}(280)}{\epsilon_{pro}(266)\, \epsilon_{nuc}(280) - \epsilon_{pro}(280)\, \epsilon_{nuc}(266)} \end{align*} Mass maps in femtograms per pixel are then computed via $n \times$ pixel area $\times$ molar mass.

5. System Performance

The DART platform achieves distinctive performance in terms of resolution, sensitivity, and operational breadth:

Parameter	Performance/Value	Notes
Lateral resolution	435 nm (266/280 nm LED), 308 nm (405 nm laser)	Resolved USAF target, Supplementary Fig. S5
Field of view	~2 cm² (~40 × 40 mm²)	Single scan
Depth of field	0.2–2 mm (mm-scale DOF)	Digital refocusing enabled
Molecular sensitivity	~50 fg/pixel (nucleic acid), ~800 fg/pixel (protein)	Figure 1d, f
Contrast improvement	9×–19× versus conventional 20×/0.75-NA brightfield	Figure 6c

A plausible implication is that DART can bridge a regime between diffraction-limited optical microscopy and classical brightfield cytometry, simultaneously capturing molecular content and mesoscale tissue context.

6. Biological Applications and Virtual Staining

DART enables quantitative, label-free imaging across diverse biological specimens:

• Cytopathology: For lung FNA smears, DART generates cm² topographic and molecular contrast maps, revealing cytoplasmic and nuclear features without stains.
• Neural tissue: Unstained mouse amygdala sections show amplitude contrast delineating GABAergic neuronal populations, undetected in standard brightfield/phase modalities.
• Blood cytometry: Permits measurement of physical and molecular features per cell (area, nuclear-cytoplasmic ratio, roundness), separating leukocyte subtypes in situ.
• Cell cultures/epithelia: DART produces differential spectroscopic maps of fixed HEK-293 cells and buccal smear samples.

The extracted $n_{pro}(x, y)$ and $n_{nuc}(x, y)$ maps support two virtual staining methods:

• Fluorescence-style: nucleic acids map to blue, proteins to purple.
• H&E-style: nucleic acids rendered hematoxylin-blue, proteins eosin-pink, with background offset in green to produce white. Crucially, stain intensities directly represent measured molecular mass, minimizing “black-box” mapping and yielding physics-explainable virtual stains.

7. Limitations, Advancements, and Future Directions

DART exhibits several limitations. Partial temporal coherence of DUV LEDs necessitates multi-mode modeling, increasing data acquisition time (~40 s per wavelength for 400 patterns). Current reconstructions are moderate speed (minutes, MATLAB implementation), and only two DUV excitation wavelengths are established, restricting specificity beyond proteins and nucleic acids.

Planned enhancements include multiplexed multi-wavelength (simultaneous LED) operation with spectral unmixing, additional virtual states (e.g., wavelength or mask perturbations) to trap non-modeled artifacts, GPU/deep-learning–driven reconstructions (implicit neural representations), angle-varied illumination for 3D multiplexing and further resolution increase (Fourier aperture synthesis), and metasurface-encoded coded masks for polarization or additional property extraction. Extensions toward near-IR vibrational contrast for chemical mapping are also proposed, albeit with trade-offs in resolution.

Anticipated future applications include ptycho-endoscopic probes for in-vivo DUV molecular imaging, space-flight platforms for astrobiology and astronaut health, and field-portable, point-of-care diagnostics—particularly in resource-constrained environments.

DART’s combination of lensless imaging, DUV spectroscopic mapping, virtual artifact-nullification, and explainable virtual staining enables high-resolution, label-free molecular analysis in a handheld device, with implications across biomedical diagnostics, clinical pathology, and fundamental biological research.