Optically Detected Magnetic Resonance Imaging

• ODMRI is a spatially resolved imaging modality that employs optical spin readout to map physical fields such as magnetic field, temperature, strain, and pressure.
• It utilizes varied spatial encoding strategies, including pulsed-gradient and wide-field imaging, to balance trade-offs between sensitivity, spatial resolution, and temporal performance.
• Recent advances integrate high-speed detection, robust microwave delivery, and improved NV center engineering to enable dynamic and extreme-condition imaging applications.

Optically detected magnetic resonance imaging (ODMRI) denotes the use of optically detected magnetic resonance as a spatially resolved measurement modality. In the strict sense introduced for nitrogen-vacancy (NV) centers in diamond, it combines ODMR detection with pulsed magnetic field gradients and Fourier spatial encoding; in the broader literature, the term also covers wide-field, scanning, volumetric, and discrete-sensor schemes in which spatial information is supplied optically and spin information is encoded in the ODMR spectrum. Across these variants, ODMRI has been used for magnetic-field mapping and, more generally, for temperature, strain, and pressure readout in room-temperature solid-state platforms (Blank et al., 2014, Blankenship et al., 23 Feb 2025, Ohkuma et al., 18 Jul 2025).

1. Conceptual scope and nomenclature

The original formulation of ODMRI was explicitly MRI-like: ODMR detection was combined with short pulsed magnetic field gradients so that the image was acquired in parallel from all parts of the sample, with a well-defined three-dimensional point-spread function and without loss of spectroscopic information (Blank et al., 2014). Subsequent usage broadened the term. In NV-ensemble magnetometry, ODMRI often denotes wide-field or confocal imaging in which each pixel carries an ODMR spectrum and spatial localization arises from direct optical imaging rather than from gradient or phase encoding. In more recent work, the term has been extended further to sparse multi-point mapping with discrete sensors and to volumetric optical scans of three-dimensional microstructures (Blankenship et al., 23 Feb 2025, Ohkuma et al., 18 Jul 2025).

This broader usage is a recurring source of terminological ambiguity. One line of work reserves “ODMRI” for gradient-encoded magnetic resonance imaging in the conventional MRI sense; another uses it for any spatially resolved ODMR map. A related misconception is that all ODMRI implementations use magnetic field gradients. Several of the best-known NV demonstrations state the opposite: there are no magnetic field gradients or phase-encoding schemes, and the spatial information arises purely from direct optical imaging of the sensing layer (Steinert et al., 2010).

Modality	Spatial encoding	Representative work
Pulsed-gradient ODMRI	Short pulsed magnetic field gradients and Fourier reconstruction	(Blank et al., 2014)
Wide-field NV ODMR imaging	Direct optical imaging of an NV layer; one ODMR spectrum per pixel	(Steinert et al., 2010)
Dynamic lock-in magnetic microscopy	Per-pixel frequency-modulated ODMR with lock-in camera readout	(Parashar et al., 2021)
Volumetric ODMR in microarchitectures	Confocal optical scanning through a 3D structure	(Blankenship et al., 23 Feb 2025)
Discrete-sensor ODMRI under pressure	Multiple microdiamonds acting as localized sensing sites	(Ohkuma et al., 18 Jul 2025)
Integrated infrared magnetic camera	Intersecting pump and IR beams defining an on-chip pixel matrix	(Bopp et al., 2023)
Raster-scanned photodetection	AOD-steered focused beam with single-photodiode readout	(Troise et al., 1 Dec 2025)

2. Spin-physical basis of ODMR imaging

For NV centers, the operative defect is a spin-triplet ground state with sublevels $m_s=0,\pm1$. In zero magnetic field, the dominant splitting is the zero-field splitting $D \approx 2.87\,\text{GHz}$, and under optical excitation the defect is pumped predominantly into $m_s=0$, which is the bright state. The $m_s=\pm1$ manifold is darker because of a higher probability of intersystem crossing through a non-radiative singlet pathway; the corresponding optical spin contrast is $\sim 30\%$ in the shallow-ensemble wide-field implementation (Steinert et al., 2010). Applying microwaves resonant with the spin transitions transfers population from $m_s=0$ into $m_s=\pm1$, producing the fluorescence dips that define the ODMR spectrum.

A standard effective Hamiltonian for an NV ensemble is

$$H = h D S_z^2 + h E (S_x^2 - S_y^2) + g \mu_B \mathbf{B}\cdot \mathbf{S},$$

with $D \approx 2.87\,\text{GHz}$, an effective transverse parameter $E$, and the Zeeman term $D \approx 2.87\,\text{GHz}$0 lifting the $D \approx 2.87\,\text{GHz}$1 degeneracy (Parashar et al., 2021). In the shallow-ensemble vector-imaging formulation, each crystallographic NV orientation $D \approx 2.87\,\text{GHz}$2 is modeled by an orientation-dependent Hamiltonian $D \approx 2.87\,\text{GHz}$3, and the external field produces two ODMR resonances per orientation, for up to eight resonances in an ensemble spectrum because diamond contains four NV orientations (Steinert et al., 2010). The resonance frequencies therefore encode field magnitude and direction, while the optical signal provides the readout channel that makes imaging practical.

This spin physics generalizes beyond magnetic fields. In high-pressure microdiamond sensing, the zero-field splitting itself is used as the encoded quantity, with

$D \approx 2.87\,\text{GHz}$4

where $D \approx 2.87\,\text{GHz}$5 and $D \approx 2.87\,\text{GHz}$6, so that local pressure is reconstructed from the ODMR frequency shift (Ohkuma et al., 18 Jul 2025). In nanodiamond-coated three-dimensional gyroids, the temperature dependence of the zero-field splitting is used instead, with $D \approx 2.87\,\text{GHz}$7 near room temperature (Blankenship et al., 23 Feb 2025). In all cases, ODMRI hinges on the same principle: spin resonance is detected optically, and the local resonance parameters are converted into a spatial map of a physical field.

3. Spatial encoding architectures

The pulsed-gradient implementation established the strict MRI-like version of ODMRI. In that scheme, a Hahn echo is combined with phase-encoding gradients, and a third microwave pulse at the echo time rotates transverse magnetization onto the $D \approx 2.87\,\text{GHz}$8-axis for optical detection. Because ODMR natively detects spin populations rather than transverse magnetization, the experiment is repeated with detection pulses along two orthogonal phases to recover the complex spin echo, after which a Fourier transform yields the spatial image. The proof-of-principle demonstration produced a $D \approx 2.87\,\text{GHz}$9 pixel image with spatial resolution of $m_s=0$0 in one direction and $m_s=0$1 in the other (Blank et al., 2014).

The best-known alternative is wide-field NV-ensemble imaging. A shallow two-dimensional layer of NV centers is illuminated over a $m_s=0$2 field of view, fluorescence is imaged onto a $m_s=0$3 CCD, and the microwave frequency is swept so that one image is acquired per frequency step. The result is a three-dimensional data cube $m_s=0$4, from which an ODMR spectrum is reconstructed for every pixel. This implementation uses no magnetic field gradients, no structured illumination, and no $m_s=0$5-space encoding; the spatial resolution is set by the optical system and the NV-layer depth (Steinert et al., 2010).

Later work diversified the encoding and readout strategies. Per-pixel lock-in detection with frequency-modulated ODMR replaced static wide-field acquisition times of few to several minutes per magnetic-field image by 50–200 fps dynamic magnetic field microscopy, using a lock-in camera that demodulates the fluorescence derivative signal at each pixel (Parashar et al., 2021). A different serial architecture replaced the camera by a focused spot raster-scanned with an acousto-optic deflector and fluorescence read out on a single photodetector; in this geometry, spatial information is time-multiplexed, and the system operates in a quasi-continuous-wave ODMR regime in which short optical pump pulses provide spin readout and repolarization while the microwave field remains continuously on resonance (Troise et al., 1 Dec 2025).

Integrated and non-planar variants further widened the concept. A diamond-on-chip infrared absorption camera used perpendicularly intersecting 532 nm pump beams and 1042 nm probe beams to define a $m_s=0$6 pixel matrix in the diamond, with balanced photodetection of infrared absorption rather than fluorescence (Bopp et al., 2023). A three-dimensional microporous gyroid coated with millions of NV-containing nanodiamonds used refractive-index-matched confocal microscopy to obtain voxelized ODMR spectra from $m_s=0$7 areas across a cross-section; the authors explicitly noted that this “ODMRI” is optical scanning of ODMR contrast in three dimensions rather than gradient-encoded MRI (Blankenship et al., 23 Feb 2025).

4. Vector reconstruction and representative magnetic-imaging demonstrations

A canonical ODMRI demonstration is the wide-field NV magnetometer based on a shallow implanted sensor sheet under a pair of current-carrying micro-wires. For each pixel, the ODMR spectrum is fitted to extract the resonance positions $m_s=0$8, and a numerical least-squares procedure then determines the field vector $m_s=0$9 from the orientation-dependent Hamiltonians. This yielded full two-dimensional vector imaging over a $m_s=\pm1$0 field of view, with a magnetic sensitivity of $m_s=\pm1$1, spatial resolution of 250 nm, and room-temperature operation (Steinert et al., 2010).

Vector reconstruction is not unique unless symmetry is broken. Because the NV center has $m_s=\pm1$2 symmetry and four equivalent crystallographic orientations, up to 24 vector orientations can fit the same set of projections when the field is not aligned to an NV axis. The same work therefore emphasized two routes to a unique solution: applying a tiny bias field along one preferred axis and monitoring the corresponding perturbation, or using a priori knowledge of the field geometry to select the physically correct branch (Steinert et al., 2010). This issue remains central in ensemble ODMRI whenever full vector magnetometry is attempted.

Time-resolved magnetic imaging extended ODMRI beyond static fields. Per-pixel lock-in detection of frequency-modulated NV photoluminescence enabled magnetic microscopy of micro-wires and microcoils with frame rates from 50 to 208 fps, a field of view of about $m_s=\pm1$3, per-pixel spatial resolution of $m_s=\pm1$4–$m_s=\pm1$5, and a median per-pixel sensitivity of $m_s=\pm1$6. The same platform captured periodic and arbitrary waveform currents, demonstrating that ODMRI can probe genuinely dynamic microscale field processes rather than only time-averaged textures (Parashar et al., 2021).

Integrated architectures reached a different operating point. The infrared-absorption magnetic camera reconstructed the position of an electromagnet in space from a $m_s=\pm1$7 pixel map, with reconstructed position uncertainties $m_s=\pm1$8. In that device, the best pixel reached an rms magnetic sensitivity of $m_s=\pm1$9, the worst $\sim 30\%$0, and the pixel-average was $\sim 30\%$1 (Bopp et al., 2023). The AOD-based serial method, by contrast, targeted weak dynamic signals in conductive media and achieved sub-millisecond temporal resolution with $\sim 30\%$2 per pixel sensitivity, illustrating a trade-off between dense parallelization and low-noise serial photodetection (Troise et al., 1 Dec 2025).

5. Non-planar, volumetric, and extreme-condition ODMRI

ODMRI is not restricted to planar magnetic mapping near a diamond surface. In a nanopolycrystalline diamond anvil cell, several NV-containing microdiamonds of 15–25 $\sim 30\%$3 diameter were sealed in the sample chamber and read out by continuous-wave ODMR up to pressures exceeding 20 GPa. Because the microdiamonds occupied different positions, their resonance frequencies differed, yielding localized pressure values of $\sim 30\%$4, $\sim 30\%$5, and $\sim 30\%$6 for three sensing sites at the same macroscopic load. The work explicitly described this as a primitive, discrete-sensor form of ODMRI: each microdiamond acts as a coarse pixel in a pressure map, and the same principle is intended to extend to magnetic imaging under concurrent high-pressure and high-temperature conditions (Ohkuma et al., 18 Jul 2025).

A distinct expansion of the field is three-dimensional microarchitecture sensing. A $\sim 30\%$7 triply periodic minimal surface gyroid was fabricated by multiphoton lithography and surface-functionalized with millions of NV-containing nanodiamonds. Refractive-index-matched confocal imaging enabled volumetric optical access, and ODMR spectra were extracted from $\sim 30\%$8 regions across a cross-section. In wide-field ensemble operation, the same platform provided temperature sensing with an average sensitivity of $\sim 30\%$9 at 5 mW excitation power (Blankenship et al., 23 Feb 2025).

These extensions clarify that ODMRI is now applied across several geometrical regimes: thin sensor sheets for near-field vector magnetometry, sparse distributed probes in confined high-pressure chambers, and volumetric porous scaffolds for three-dimensional thermometry and, plausibly, multimodal sensing. A plausible implication is that the defining feature of ODMRI is no longer a specific spatial-encoding mechanism but the combination of optical spin readout with spatially resolved reconstruction of a resonance-derived observable.

6. Performance trade-offs, instrumentation, and outlook

The central trade-off in ODMRI is between sensitivity, spatial resolution, field of view, and temporal bandwidth. Wide-field NV ensembles improve sensitivity as $m_s=0$0 and offer full vector reconstruction from all four orientations, but they are diffraction-limited and, in static implementations, can require long acquisition times. Scanning-probe NV magnetometry offers tens-of-nanometer spatial resolution but is slow and usually measures only one NV orientation at a time. Wide-field ODMRI is therefore complementary: it sacrifices ultimate spatial resolution for speed, field of view, and multiplexed vector information (Steinert et al., 2010).

Temporal performance has been a particular bottleneck. Conventional wide-field NV ODMR imaging was described as requiring few to several minutes of acquisition to get a single magnetic field image, effectively restricting the technique to temporally static phenomena. Frequency-modulated lock-in imaging and AOD-based quasi-continuous-wave scanning were developed precisely to overcome this limitation, and now support 50–200 fps or sub-millisecond temporal resolution, respectively (Parashar et al., 2021, Troise et al., 1 Dec 2025). The price is that high-speed operation is typically performed at fixed microwave frequency on the ODMR slope, so quantitative performance depends strongly on linewidth uniformity, bias-field stability, and calibration of the local slope.

Microwave delivery is itself a limiting and enabling technology. A planar ring antenna designed for ODMR of NV centers was engineered to resonate around 2.87 GHz with a bandwidth of 400 MHz and a spatially uniform microwave field within a 1-mm-diameter center hole, explicitly to support magnetic-field imaging over a wide spatial range. Because the antenna is located beneath the diamond, it preserves optical access to the top surface, making it compatible with both scanning and wide-field ODMRI geometries (Sasaki et al., 2016).

The principal limitations identified across the literature are also consistent. In ensemble NV layers, higher density increases fluorescence and shortens exposure time but broadens ODMR lines through nitrogen-related dipolar dephasing; lower density narrows the lines but reduces photon flux and therefore demands longer integration (Steinert et al., 2010). In integrated infrared cameras, present sensitivities are far from the single-pixel benchmark because of limited pump power, imperfect beam overlap, and low singlet-absorption contrast, but proposed remedies include non-absorbing adhesives, localized NV creation, infrared cavities with finesse $m_s=0$1, and magnetic flux concentrators (Bopp et al., 2023). In extreme-condition sensing, anisotropic strain, non-uniform pressure, and microdiamond motion complicate interpretation and reproducibility (Ohkuma et al., 18 Jul 2025).

Future development is therefore likely to proceed along several convergent lines already stated in the literature: improved NV engineering and annealing to reduce paramagnetic impurities; coherent spin manipulation techniques such as Ramsey interferometry, spin echoes, and dynamical decoupling to exploit longer phase-memory times; brighter excitation and higher collection efficiency to increase photon number $m_s=0$2; faster or lower-noise cameras and photodetectors to improve temporal resolution; algorithmic regularization and bias fields to remove vector ambiguities; and increasingly integrated microwave and photonic hardware for robust, adjustment-free devices (Steinert et al., 2010, Bopp et al., 2023). The resulting picture is of ODMRI as a family of methods rather than a single protocol: a shared spin-optical principle implemented through multiple spatial-encoding architectures, each optimized for a different combination of scale, speed, and environment.