Tomosynthesis

Tomosynthesis, also known as digital tomosynthesis (DTS), is a limited-angle tomographic imaging technique that uses conventional X-ray equipment to acquire a series of low-dose projection images as the X-ray tube moves along an arc, typically spanning less than 40 degrees, which are then reconstructed computationally to produce high-resolution sectional images with reduced out-of-plane blurring.^[1] This method provides pseudo-three-dimensional visualization of anatomical structures at radiation doses comparable to or lower than conventional two-dimensional radiography, while being more cost-effective and less complex than full computed tomography (CT).^[1] The core principle of tomosynthesis involves capturing multiple two-dimensional projections—often 11 to 71 images depending on the application—and applying reconstruction algorithms such as filtered backprojection (FBP), shift-and-add (SAA), or iterative methods to synthesize in-focus planes while suppressing structures outside the plane of interest.^[1] Originating from early 20th-century concepts of planar tomography introduced in the 1930s by Ziedses des Plantes, the term "tomosynthesis" was coined in 1972 by Allan Grant, with modern digital implementations advancing significantly in the late 1990s thanks to flat-panel detector technology.^[1] Compared to traditional radiography, tomosynthesis significantly enhances lesion detectability by minimizing superimposition of overlapping tissues; for instance, it improves sensitivity for pulmonary nodules from 22% in posteroanterior chest radiography to up to 70%.^[1] Tomosynthesis has diverse clinical applications, most prominently in breast imaging where digital breast tomosynthesis (DBT), FDA-approved in 2011, generates 3D mammographic images from X-ray projections taken at multiple angles during a single compression, aiding in the detection of breast cancer and reducing false positives.^[2][3] Other key uses include chest imaging for evaluating pulmonary nodules, infections, and rib fractures with improved visualization over standard radiographs; orthopedic assessments for joint and bone abnormalities; dental imaging for clearer views of teeth and jaws; and interventional procedures in radiation oncology for precise tumor localization.^[1] As of 2025, it is widely used in breast and chest imaging, with established roles in musculoskeletal, dental, and industrial applications. Advantages include higher diagnostic accuracy, such as an approximately 25% to 50% relative increase in cancer detection rates in mammography screening, alongside lower recall rates and reduced patient radiation exposure relative to CT equivalents.^[4][5] Recent advancements include AI integration for further enhancements in detection performance.^[6]

Fundamentals

Definition and Basic Principles

Tomosynthesis is an imaging technique that enables the creation of high-resolution, three-dimensional (3D) representations of an object using a limited number of two-dimensional (2D) X-ray projections acquired over a restricted angular range, typically resulting in radiation doses comparable to those of conventional radiography.^[7] Unlike full computed tomography (CT), which requires projections over 180° or more for complete volumetric reconstruction, tomosynthesis employs limited-angle tomography to produce a stack of thin image slices (often 1 mm thick) that separate overlapping anatomical structures and reduce superimposition artifacts.^[8] This method is particularly valuable in applications such as breast imaging, where it improves lesion detection by providing depth-resolved views without the need for excessive radiation exposure.^[9] The basic principle of tomosynthesis revolves around the acquisition of multiple low-dose projection images as the X-ray tube moves along an arc or linear path relative to the stationary detector and the imaged object. In a typical setup, the tube sweeps through an angle of 15° to 50° (depending on the system), capturing 10 to 30 projections during a single exposure sequence that lasts 10 to 20 seconds.^[4] This geometry ensures that in-plane resolution remains high (similar to 2D radiography) while introducing depth discrimination by blurring structures outside the focal plane, a concept inherited from analog tomography but enhanced digitally.^[8] The total radiation dose is distributed across the projections, often equaling that of a standard 2D mammogram, making it suitable for screening.^[9] Image reconstruction in tomosynthesis involves algorithmic processing of the projection data to generate the 3D slice stack, addressing the underdetermined nature of limited-angle data. Common methods include filtered back-projection (FBP), which applies ramp and apodizing filters to the projections before back-projecting them onto parallel planes, and iterative techniques such as simultaneous algebraic reconstruction technique (SART) or maximum-likelihood expectation maximization (MLEM), which iteratively refine the estimate to minimize artifacts and noise.^[8] These approaches yield non-isotropic resolution, with superior in-plane detail but coarser depth resolution compared to CT, prioritizing clinical utility in reducing anatomical noise for better visualization of subtle features like microcalcifications.^[4]

Data Acquisition Process

The data acquisition process in tomosynthesis entails the capture of multiple low-dose, two-dimensional X-ray projection images of the target anatomy from a series of discrete angles within a limited angular range, typically spanning 15° to 60°. This approach, known as limited-angle tomography, enables the generation of quasi-three-dimensional images by reducing the superposition of structures that occurs in conventional projection radiography, while maintaining radiation doses comparable to standard two-view imaging.^[10] Unlike full computed tomography, which requires projections over 180° or more, tomosynthesis prioritizes speed and dose efficiency, with acquisition times generally lasting 3 to 25 seconds.^[10] In most clinical systems, the X-ray tube is mounted on a rotating gantry that arcs around the stationary digital detector, with the patient positioned such that the region of interest—such as the breast in digital breast tomosynthesis—is compressed between a paddle and the detector to ensure uniform thickness and minimize motion artifacts. The tube motion follows one of two schemes: step-and-shoot, in which the tube halts at each angle for a brief exposure (e.g., 100-200 ms per projection), or continuous motion, where the tube sweeps steadily while the X-ray beam remains active, necessitating post-acquisition corrections for potential blur from tube and detector movement.^[10] For example, in breast imaging, projections are acquired in cranio-caudal or medio-lateral oblique views, with the angular distribution often symmetric around the zero-degree (central) position to optimize depth resolution.^[11] The number of projections acquired ranges from 9 to 25, selected to balance spatial resolution, noise levels, and radiation exposure; research indicates that 15 to 20 projections suffice for effective separation of in-plane and out-of-plane structures in typical clinical scenarios. Commercial implementations illustrate this variability: Hologic's Selenia Dimensions system uses 15 projections over a 15° arc in about 3.7 seconds, GE's Senographe Pristina employs 9 projections across 25°, and Siemens' MAMMOMAT Inspiration captures 25 projections over 50°. Exposure parameters, including tube voltage (typically 25-35 kVp) and current-time product, are distributed either uniformly or variably across angles to maintain consistent image quality, with total mean glandular dose for breast tomosynthesis approximating that of digital mammography (e.g., 3-4 mGy per view for a 5 cm compressed breast).^[10] X-ray sources commonly feature tungsten anodes with aluminum or rhodium filtration (0.05-0.7 mm thickness) to produce a spectrum optimized for soft-tissue contrast, while detectors are flat-panel arrays with pixel pitches of 50-100 μm, often using direct-conversion amorphous selenium for high detective quantum efficiency and spatial resolution exceeding 5 line pairs per millimeter. In non-breast applications, such as chest or dental tomosynthesis, similar principles apply but with adjusted geometries to accommodate larger fields of view or upright patient positioning. Scatter radiation, which can degrade contrast, is managed primarily through software-based corrections (e.g., Monte Carlo simulations) rather than anti-scatter grids, as the limited angular range inherently reduces scatter-to-primary ratios compared to wider-angle techniques.^[10] Optimization studies emphasize trade-offs in geometry: narrower arcs (e.g., 15°-25°) preserve higher in-plane resolution but limit depth discrimination, whereas wider arcs (45°-60°) enhance slice separation at the cost of increased acquisition time and potential artifacts from off-axis magnification. These parameters are tailored to specific applications, with seminal simulations demonstrating that a 50° arc with 17 projections yields optimal lesion detectability in breast tissue models.^[10]

Historical Development

Early Concepts and Analog Tomosynthesis

The early concepts of tomosynthesis emerged in the mid-1930s as an extension of conventional planigraphy, aiming to generate multiple tomographic planes from a limited set of projections to reduce superimposition of structures in radiographic imaging. Dutch radiologist Bernard George Ziedses des Plantes is credited with pioneering this approach in 1934, when he described the geometric principles for reconstructing arbitrary planes using shift-and-add superposition of projections acquired during linear tube motion, building on his earlier work in planigraphy where the X-ray tube and film moved synchronously to focus on a single plane.^[12] Independently, in 1936, American medical doctor Julius Kaufman patented a similar device in the United States for multi-plane imaging through stepped tube movements and film exposures, enabling the synthesis of several sections from discrete angular views, though his contributions received less recognition than des Plantes'.^[12] These analog methods relied on mechanical linkages to coordinate tube displacement—typically linear or arc-like paths of 10–30 degrees—with intermittent X-ray pulses, capturing projections on stacked films via rapid changers to minimize motion blur.^[13] By the late 1930s, practical implementations appeared, such as a respiratory-gated chest tomosynthesis system that synchronized tube motion with patient breathing to acquire projections over a limited angle, producing blurred out-of-plane artifacts while sharpening in-focus layers through manual or optical re-projection.^[14] During the 1940s and 1950s, analog systems evolved with multi-film cassette changers, allowing 10–20 exposures in seconds as the tube traversed predefined paths, often linear for simplicity; these were applied in orthopedics and chest imaging to visualize overlapping bones or lungs.^[12] Reconstruction in these era's devices involved analog back-projection: films were physically shifted relative to each other under diffuse illumination and superimposed photographically, a labor-intensive process that approximated 3D sections but suffered from low contrast and artifacts due to uneven blurring across depths.^[15] The term "tomosynthesis" was formally coined in 1972 by D.G. Grant in a seminal paper, defining it as a technique for synthesizing tomographic slices from a finite set of 2D projections via computational or optical shift-and-add algorithms, reviving interest in limited-angle tomography as an alternative to full computed tomography (CT).^[15] Grant's work emphasized digital potential but built on analog foundations, demonstrating reconstructions from 9–15 projections over 30–60 degrees with resolution comparable to single-plane tomography (approximately 1–2 mm in-plane).^[16] In the 1970s, advanced analog variants included coded-aperture systems developed by Klotz and Weiss at Philips, which used modulated exposure patterns on film to enable deconvolution-like reconstruction of multiple planes, improving signal-to-noise ratios in chest applications without digital processing.^[12] These film-based methods, while innovative, were limited by mechanical precision, radiation dose from multiple exposures (often 2–5 times conventional radiography), and reconstruction complexity, paving the way for digital transitions in the 1980s.^[14]

Digital Era and Clinical Adoption

The digital era of tomosynthesis emerged in the late 1990s, enabled by the development of high-resolution digital flat-panel detectors that facilitated the capture of multiple low-dose projections for three-dimensional reconstruction. This shift from analog film-based systems to digital acquisition addressed longstanding challenges such as motion artifacts and limited image quality in earlier tomosynthesis techniques. A pivotal advancement occurred in breast imaging with the 1997 study by Niklason et al., which first demonstrated the feasibility of digital breast tomosynthesis (DBT) using a full-field digital mammography system to acquire and reconstruct sectional images from mastectomy specimens and volunteers, revealing reduced superimposition of breast tissue compared to two-dimensional mammography. Concurrently, broader applications were explored; for instance, a 2003 review by Dobbins et al. outlined the principles and potential of digital x-ray tomosynthesis for general radiography, including chest imaging, where it could generate pseudo-slice images from a single tube sweep to improve visualization of overlapping structures.^[17] Commercial systems proliferated in the mid-2000s, driven by collaborations between academic researchers and industry. GE Healthcare launched the VolumeRAD digital tomosynthesis option for its Definium 8000 radiography system in 2006, targeting chest and musculoskeletal applications with flat-panel detectors for rapid, low-radiation imaging. In breast imaging, regulatory milestones accelerated progress: the U.S. Food and Drug Administration (FDA) approved Hologic's Selenia Dimensions DBT system in February 2011 as the first clinically deployable device, followed by GE Healthcare's SenoClaire in 2014.^[18][19] These approvals were bolstered by evidence from prospective trials, such as the 2013 interim results from the Oslo Tomosynthesis Screening Trial, which found that adding DBT to digital mammography increased invasive cancer detection by 27% and improved specificity by reducing false-positive recalls.^[20] Clinical adoption has varied by application, with breast screening leading due to robust evidence of efficacy. In the United States, DBT use in screening mammography surged from 12.9% of examinations in early 2015 to 43.2% by late 2017, surpassing 50% in 42% of hospital referral regions, influenced by higher detection rates and socioeconomic factors like regional income levels. By 2024, DBT had been adopted by over 90% of breast imaging centers in the United States, becoming the standard for screening mammography.^[21][22] For chest imaging, adoption began with installations in specialized centers post-2006, where interim NIH-sponsored trials in 2008 demonstrated 2- to 3-fold greater sensitivity for small lung nodules (4-14 mm) compared to conventional posteroanterior radiography, positioning it as a low-dose alternative to computed tomography in resource-limited settings. However, chest tomosynthesis remains niche, with slower integration due to CT's established role. In musculoskeletal imaging, digital tomosynthesis enhances detection of subtle fractures and orthopedic hardware complications, as shown in studies since the late 2000s, but it is typically used adjunctively rather than routinely, with adoption confined to high-volume trauma centers.^[23][24]

Image Reconstruction

Core Reconstruction Methods

Tomosynthesis reconstruction involves solving an ill-posed inverse problem to generate three-dimensional images from a limited set of two-dimensional projections acquired over a restricted angular range, typically 15–50 degrees, which leads to challenges such as limited angular sampling and artifacts like blurring from out-of-plane structures.^[25] Core methods are broadly categorized into analytical approaches, which provide rapid solutions through direct inversion, and iterative techniques, which refine estimates progressively to improve image quality at the cost of computation time. These methods adapt principles from computed tomography but account for the incomplete data in tomosynthesis, often incorporating regularization to mitigate artifacts.^[26] The foundational analytical method is filtered back-projection (FBP), which processes each projection by applying a ramp filter to compensate for the blurring inherent in simple back-projection, followed by summation along rays to form the 3D volume. In tomosynthesis, FBP is modified with additional filters, such as apodization to reduce high-frequency noise and slice-thickness filters to suppress out-of-plane artifacts, enabling efficient reconstruction suitable for clinical workflows.^[27] This approach offers high speed and reproducibility, allowing dose reductions of up to 50% in applications like digital breast tomosynthesis while maintaining diagnostic utility, though it can amplify noise and streak artifacts in limited-angle scenarios.^[25] Seminal implementations include those by Mertelmeier et al. for breast imaging, demonstrating its practicality in commercial systems.^[27] Iterative reconstruction techniques address FBP's limitations by modeling the imaging physics more explicitly and incorporating statistical priors, iteratively updating voxel estimates to minimize discrepancies between simulated and measured projections. Algebraic reconstruction technique (ART) updates voxels sequentially using data from one projection at a time, converging quickly but potentially introducing noise in sparse data sets.^[28] In contrast, simultaneous algebraic reconstruction technique (SART) and simultaneous iterative reconstruction technique (SIRT) process all projections concurrently in each iteration, yielding smoother images with reduced streak artifacts; SART, for instance, has shown superior in-plane resolution and fewer iterations than expectation-maximization methods in breast tomosynthesis evaluations.^[29] Another key method, maximum likelihood expectation maximization (MLEM), statistically optimizes the likelihood of observed projections under a Poisson noise model, enhancing contrast and suppressing artifacts after 10–20 iterations, as demonstrated in early tomosynthesis studies for mammography.^[30] These iterative methods, originating from foundational work by Gordon et al. on algebraic techniques and adapted by Wu et al. for limited-angle cone-beam data, enable better handling of low-dose acquisitions but require significant computational resources, often mitigated by parallel processing.^[28][30] Overall, while FBP remains the clinical standard for its efficiency, iterative approaches like SART and MLEM are increasingly adopted for their superior artifact reduction and quantitative accuracy in high-impact applications.^[26]

Advanced and Iterative Techniques

Advanced and iterative reconstruction techniques in tomosynthesis represent a significant evolution from analytical methods like filtered backprojection, addressing the inherent challenges of limited-angle acquisition geometry, such as incomplete data and amplified noise. These methods iteratively refine an initial image estimate by modeling the forward projection process and minimizing a discrepancy measure between measured and simulated projections, often incorporating statistical priors or regularization to enhance image quality. In breast tomosynthesis, iterative approaches have demonstrated superior contrast-to-noise ratios and reduced out-of-plane artifacts compared to direct methods, improving lesion detectability.^[26] Algebraic iterative reconstruction techniques, such as the Algebraic Reconstruction Technique (ART) and its parallel variant, Simultaneous Algebraic Reconstruction Technique (SART), solve a system of linear equations derived from the ray-tracing model of projections. ART updates pixel values sequentially using projections from individual angles, while SART processes all projections simultaneously per iteration, leading to faster convergence and more uniform noise distribution. In a comparative study for breast tomosynthesis, SART initialized with backprojection results achieved higher in-plane resolution and lower artifacts than filtered backprojection, particularly for microcalcifications, with reconstruction times suitable for clinical use after 5-10 iterations.^[31] Statistical iterative methods, including Maximum Likelihood Expectation Maximization (MLEM) and Ordered Subset Expectation Maximization (OS-EM), incorporate Poisson noise statistics from X-ray photon counting to produce unbiased estimates. MLEM alternates between forward projection (expectation) and backprojection (maximization) steps, balancing low- and high-frequency details effectively in limited-angle scenarios. For digital breast tomosynthesis, Wu et al. applied MLEM to cone-beam projections, yielding images with improved signal-to-noise ratios (up to 20% better than backprojection) and enhanced visualization of subtle masses, though requiring acceleration techniques like ordered subsets to reduce computation from hours to minutes. Further advancements integrate model-based iterative reconstruction (MBIR) frameworks that explicitly account for the system geometry, detector response, and scatter, often with regularization terms like total variation (TV) to suppress noise while preserving edges. In digital breast tomosynthesis, a TV-regularized iterative algorithm has shown reduced out-of-focus blur in phantom studies compared to unregularized methods, enabling clearer separation of overlapping structures. GPU-accelerated MBIR variants have made these computationally intensive techniques feasible for real-time clinical workflows, achieving reconstruction speeds of under 1 minute per volume while maintaining high detective quantum efficiency. Recent hybrid approaches combine iterative methods with deep learning priors, such as convolutional neural networks for initialization, to accelerate convergence and improve microcalcification visibility without increasing radiation dose.^[32][33][34]

Comparisons with Other Modalities

Versus Conventional Radiography

Tomosynthesis differs fundamentally from conventional radiography in its approach to image acquisition and reconstruction. Conventional radiography produces two-dimensional projection images from one or a few X-ray angles, resulting in the superposition of anatomical structures that can obscure pathologies. In contrast, tomosynthesis acquires multiple low-dose projections over a limited angular range (typically 15–50 degrees) and reconstructs them into a series of thin, pseudo-three-dimensional slices, enabling better separation of overlapping tissues.^[12][9] This limited-angle tomography provides depth information without the full 360-degree rotation required for computed tomography (CT), bridging the gap between 2D and full 3D imaging.^[35] A primary advantage of tomosynthesis over conventional radiography is enhanced lesion detection due to reduced anatomical overlap and improved contrast resolution. In breast imaging, digital breast tomosynthesis (DBT) increases cancer detection rates by 1.6–2.0 per 1,000 women screened and reduces recall rates by 15–37% compared to full-field digital mammography, particularly benefiting women with dense breasts.^[36] For chest applications, digital chest tomosynthesis (DCT) demonstrates superior sensitivity (85% vs. 47%) and specificity (95% vs. 37%) in detecting pulmonary nodules, with a per-lesion detection ratio of nearly 3:1, by minimizing superimposition from ribs and vessels.^[37] In musculoskeletal imaging, tomosynthesis improves visualization of fractures, osteonecrosis, and hardware complications, outperforming radiography in detecting subtle bone changes like demineralization near implants or wrist fractures post-fixation.^[38][39] Additionally, acquisition time remains similar to conventional radiography (under 10 seconds), and radiation doses are comparable or slightly higher (e.g., 0.1–0.2 mSv for DCT vs. 0.06–0.10 mSv for chest radiography), making it a practical upgrade.^[35][40] Despite these benefits, tomosynthesis has limitations relative to conventional radiography. Its depth resolution is shallower than CT's, potentially complicating assessment of subpleural or deep structures, and it may introduce motion artifacts from breathing or heart movement in chest exams.^[35] In breast imaging, DBT can miss microcalcifications due to slice thickness and requires longer reading times (up to 2.3 times that of 2D mammography).^[36] Radiation exposure is higher than a single 2D projection, though synthetic 2D images can mitigate this by eliminating the need for separate full-dose mammograms.^[9] Artifacts from high-density structures, such as calcifications mimicking nodules, also pose challenges not seen in simple radiography.^[35] Overall, while tomosynthesis enhances diagnostic accuracy in overlap-prone areas, it complements rather than fully replaces conventional radiography for initial low-cost screening.^[40]

Versus Computed Tomography

Tomosynthesis and computed tomography (CT) both provide three-dimensional imaging capabilities but differ fundamentally in acquisition geometry and clinical utility. Tomosynthesis acquires a limited set of projection images over a small angular range, typically 15–50 degrees, to reconstruct pseudo-3D slices, whereas CT uses a full 180–360-degree rotation for comprehensive volumetric data. This limited-angle approach in tomosynthesis results in lower radiation exposure compared to CT, making it suitable for screening applications where dose reduction is prioritized.^[41][42] Radiation dose is a primary differentiator, with tomosynthesis delivering significantly less exposure than even low-dose CT protocols. For chest imaging, effective doses for digital tomosynthesis range from 0.08–2.23 mSv, approximately half or less of low-dose CT's 1.1–4.1 mSv.^[41] In pulmonary cystic fibrosis evaluation, tomosynthesis doses are about 0.08–0.12 mSv, versus over 1.1 mSv for low-dose CT, enabling more frequent monitoring without excessive cumulative risk.^[43] This dose advantage positions tomosynthesis as a cost-effective alternative for follow-up imaging, with procedure costs around 54€ compared to 161–214€ for CT.^[43] Image quality in tomosynthesis offers superior in-plane resolution for detecting fine structures like bony details or pulmonary nodules, often with fewer metal artifacts than CT, which aids postoperative assessments. However, its limited projections lead to out-of-plane artifacts and inferior depth resolution, reducing contrast sensitivity for soft-tissue differentiation compared to CT's isotropic voxels and multiplanar reconstructions. Clinically, tomosynthesis achieves comparable sensitivity to low-dose CT for detecting artificial pulmonary nodules (observer-averaged figure of merit: 0.617 for tomosynthesis vs. 0.576 for CT), and it provides better visualization of bronchiectasis or mucus plugging in cystic fibrosis than conventional radiography, though CT excels in assessing air trapping or complex anatomy.^[41][41][42] Tomosynthesis is preferred for applications requiring rapid, low-dose 3D-like imaging, such as chest screening for nodules or sacroiliitis detection, where it outperforms radiography while avoiding CT's higher dose. In contrast, CT remains the gold standard for diagnostic confirmation needing high-fidelity volumetric detail, despite its drawbacks in dose and cost. Overall, tomosynthesis bridges the gap between 2D radiography and full CT, enhancing accessibility in resource-limited settings.^[42][43][41]

Applications

Breast Imaging

Digital breast tomosynthesis (DBT), also known as three-dimensional mammography, is a specialized application of tomosynthesis technology in breast imaging that acquires multiple low-dose X-ray projections over a limited angular range, typically 15–50 degrees, to reconstruct thin-slice images of the breast. This technique addresses the primary limitation of conventional two-dimensional (2D) full-field digital mammography (FFDM) by reducing the superimposition of breast tissues, which can obscure lesions, particularly in dense breasts. DBT systems rotate the X-ray tube in an arc while the detector remains stationary, capturing 11–30 projection images that are then reconstructed using algorithms such as filtered back-projection or iterative methods to produce a series of 1-mm-thick slices.^[4][44] In clinical practice, DBT is primarily employed for breast cancer screening and diagnosis, often combined with 2D imaging or synthetic mammography (SM), where 2D-like images are generated from the DBT projections to minimize radiation exposure. The U.S. Food and Drug Administration (FDA) approved DBT in 2011 as an adjunct to FFDM, and it has since become the standard of care in many screening programs, with adoption rates exceeding 90% of U.S. facilities as of 2025.^[45] For screening, DBT improves lesion conspicuity and characterization, enabling better detection of invasive cancers and reducing false positives. Key multicenter studies, such as the Friedewald et al. trial involving over 450,000 women, demonstrated a 41% increase in invasive cancer detection rates (from 4.0 to 5.6 per 1,000 screens) and a 15% reduction in recall rates compared to 2D mammography alone. Similarly, the Oslo Tomosynthesis Screening Trial reported a cancer detection rate increase from 6.1 to 8.0 per 1,000 screens, alongside a 40% reduction in false positives.^[46] In diagnostic settings, DBT enhances the evaluation of abnormalities identified on screening, such as masses, asymmetries, and architectural distortions, often obviating the need for additional views or ultrasound. Research indicates that DBT improves diagnostic accuracy, with one study showing a 35% reduction in the requirement for supplemental imaging and a higher positive predictive value for biopsies (47.7% versus 24.9% for 2D alone). It is particularly effective for noncalcified lesions and in women with dense breasts, where sensitivity reaches 89% compared to 78% for 2D mammography, as per a meta-analysis of screening performance. The PROSPR consortium study across multiple U.S. health systems confirmed these benefits, noting a 16% decrease in recall rates and improved detection of lower-grade, prognostically favorable cancers. Guidelines from the American College of Radiology (ACR) and National Comprehensive Cancer Network (NCCN) endorse DBT for routine screening in average-risk women, especially those with dense breasts. Recent market analyses indicate continued global growth, with the DBT market projected to reach USD 10.89 billion by 2034, driven by increasing adoption and integration with AI for enhanced detection.^{[44][47][48][49]} Despite its advantages, DBT faces challenges including approximately twice the radiation dose of 2D mammography (though SM mitigates this to near-equivalent levels), longer interpretation times (about 2.8 minutes per case versus 1.9 for 2D), and larger data storage needs (200–450 MB per exam). It is less effective for microcalcifications, where 2D imaging remains superior, and its impact on breast cancer mortality has not yet been definitively established, pending results from ongoing trials like TMIST. Global adoption varies, with slower uptake outside the U.S. due to equipment costs and workflow adjustments, though evidence supports its role in improving overall screening effectiveness.^[4][46][50]

Chest Imaging

Digital tomosynthesis (DT) in chest imaging acquires a series of low-dose X-ray projections over a limited angular range, typically 15–50 degrees, to reconstruct quasi-three-dimensional images that minimize anatomical superimposition inherent in conventional two-dimensional radiography. This technique enhances the visualization of pulmonary structures, making it particularly valuable for detecting subtle abnormalities in the lungs and thorax. DT systems, often integrated with existing digital radiography equipment, allow for rapid acquisition in a single patient position, facilitating its use in routine clinical workflows.^[51] The primary clinical application of chest DT is the detection of pulmonary nodules, where it demonstrates significantly higher sensitivity than standard chest radiography. Literature reviews indicate that DT identifies approximately three times more nodules, including small ones (e.g., 5–8 mm), due to improved depth resolution and reduced overlap from overlying structures like ribs and vessels. For instance, observer studies have reported sensitivity rates for nodule detection exceeding 80% with DT, compared to around 50% with radiography alone, while maintaining comparable specificity. This makes DT a useful tool for initial screening, follow-up of known nodules, and reducing false negatives in high-risk populations, such as smokers. Beyond nodules, DT aids in characterizing hilar abnormalities, calcified lymph nodes, and interstitial lung diseases by providing better conspicuity of parenchymal patterns.^[51][52] DT also excels in evaluating skeletal and infectious pathologies in the chest. It improves the detection of rib fractures by distinguishing them from pulmonary nodules and reducing false positives caused by superimposition, with studies showing enhanced accuracy in trauma settings. For infections, such as pulmonary tuberculosis or nontuberculous mycobacterial disease, DT better delineates calcifications, consolidations, and cavitary lesions compared to plain films. Emerging applications include monitoring airway interventions, where DT assesses post-procedural complications like stent migration or bronchial stenosis with higher sensitivity than radiography; radiation doses for these exams range from 0.3–0.4 mSv, about 5–10% of a standard chest CT dose (4–8 mSv). In interventional pulmonology, DT integrates with navigational bronchoscopy for real-time guidance in sampling peripheral lung lesions, achieving diagnostic yields of 75–93% in recent studies, offering a lower-dose alternative to cone-beam CT.^[51][53][52] Despite these benefits, DT's limited angular range results in some out-of-plane blurring, which may affect axial resolution for complex anatomies, though in-plane resolution (up to 16 line pairs per cm) surpasses that of CT (around 7 lp/cm). Novel stationary DT systems, evaluated in phantoms, confirm adequate detection of 5 mm solid and 8 mm ground-glass nodules at doses as low as 0.185 mSv, positioning DT as a potential complement to low-dose CT for lung cancer screening. Overall, chest DT bridges the gap between radiography and CT, enhancing diagnostic confidence without substantially increasing radiation exposure or cost.^[54][51]

Musculoskeletal Imaging

Tomosynthesis has emerged as a valuable tool in musculoskeletal imaging, providing limited-angle tomographic images that reduce superimposition of overlapping structures compared to conventional radiography. This technique is particularly useful for evaluating high-contrast bony anatomy, such as in the detection of fractures, assessment of joint erosions, and postoperative monitoring of orthopedic hardware. By acquiring multiple low-dose projections over a limited arc (typically 15–50 degrees), tomosynthesis generates quasi-three-dimensional images with in-plane resolution superior to computed tomography (CT) in some cases, while maintaining a radiation dose 2–3 times that of radiography but significantly lower than CT.^[55][41] In fracture detection and healing evaluation, tomosynthesis excels at identifying occult fractures and cortical disruptions that are obscured on standard radiographs due to bone overlap or metallic artifacts from fixation devices. For instance, in wrist fractures, it demonstrates higher diagnostic accuracy for cortical healing (area under the curve of 0.84 versus 0.76 for radiography) and reduces hardware-related obscuration (8% of cases versus 15% for radiography). Studies have shown its sensitivity for fracture diagnosis ranging from 77% to 87%, with specificity of 76% to 82%, outperforming radiography but remaining inferior to CT (sensitivity 93–95%, specificity 86–95%). It is especially beneficial for complex areas like the scaphoid, talus, and fifth metatarsal, where it clearly delineates fracture lines and healing progress with less dependence on radiographer expertise. Additionally, tomosynthesis supports weight-bearing examinations, such as for knee osteoarthritis, allowing functional assessment of joint alignment and cartilage space under load.^[56][55][41] For inflammatory and degenerative joint diseases, tomosynthesis enhances the visualization of erosions, cysts, and osteophytes in conditions like rheumatoid arthritis, osteoarthritis, and spondyloarthritis. In rheumatoid arthritis, it achieves 77% sensitivity for detecting erosions, surpassing plain radiography by reducing overlap in small joints such as the hands and feet. Systematic reviews confirm its superiority over radiography for skeletal abnormalities in these pathologies, though results are more variable in gout. In osteoarthritis, it better characterizes joint space narrowing and subchondral changes, aiding in disease staging.^[57][55] Tomosynthesis also plays a role in tumor evaluation and postoperative imaging, where it improves detection and characterization of bone lesions like aneurysmal bone cysts and aids in assessing prosthesis loosening or osteosynthesis integrity by minimizing metallic artifacts. Compared to CT, it offers better in-plane resolution for fine bony details (up to 3.15 line pairs per millimeter) and lower radiation exposure (e.g., 0.72 mGy versus 19.8 mGy for wrist imaging), making it suitable for follow-up studies. However, limitations include reduced out-of-plane resolution and lower contrast sensitivity for soft tissues, rendering it less effective than CT for complex three-dimensional reconstructions or soft-tissue involvement. Despite these, its integration into routine musculoskeletal protocols continues to grow due to cost-effectiveness and accessibility using standard X-ray equipment.^[55][41][56]

Dental Imaging

Tomosynthesis in dental imaging is a limited-angle tomographic technique that uses multiple low-dose X-ray projections acquired over a small angular range—typically 10° to 30°—to reconstruct pseudo-three-dimensional sectional images, minimizing the overlap of teeth and bone structures that obscures details in conventional two-dimensional radiography. This method enhances the visualization of dental anatomy, including roots, alveolar bone, and interproximal spaces, making it valuable for diagnosing subtle pathologies. In dentistry, tomosynthesis is implemented primarily through intraoral systems for localized views and panoramic systems for broader jaw assessments, offering a balance between diagnostic utility and radiation efficiency compared to full cone-beam computed tomography (CBCT).^[58] Stationary intraoral tomosynthesis (sIOT) is an emerging modality that employs a fixed linear array of distributed X-ray sources, such as carbon nanotube field emitters, to capture 15 or more projections in a single exposure without mechanical motion of the patient, source, or detector. This setup spans a narrow angle, often 12°, and reconstructs image stacks using iterative algorithms like the simultaneous algebraic reconstruction technique (SART), achieving processing times of about 8 seconds for basic reconstructions or 90 seconds with total variation regularization for noise reduction. sIOT produces synthetic multi-view radiographs adjustable up to ±6° and thin slices (0.5 mm thick), improving conspicuity of interproximal caries adjacent to restorations, vertical root fractures, and periodontal defects while reducing metal streak artifacts from restorations. The total radiation dose is equivalent to traditional D-speed intraoral film, at approximately 15.75 mAs across projections, with spatial resolution matching commercial two-dimensional systems at 10 cycles per millimeter. Recent studies explore integration with AI to further enhance pathology detection in sIOT images.^[58][59] Panoramic dental tomosynthesis adapts the technique for wide-field imaging by reconstructing layered views of the entire dentition and jaws from a limited set of projections, often derived innovatively from pre-acquired CBCT data to avoid extra radiation exposure. In one method, dental arch contours are detected via maximum intensity projections and parabolic fitting, followed by rebinning of CBCT projections onto a virtual patient-specific trajectory for tomosynthesis reconstruction with auto-focusing to sharpen the focal plane. This yields higher in-plane resolution of 0.11 mm pixel size versus 0.31 mm in standard CBCT resampled images, along with diminished metal artifacts and no out-of-focus blurring, facilitating precise evaluation of edentulous areas and implant sites. Reconstruction processing takes roughly 5.5 seconds for projections and 6.2 seconds for images, with effective doses remaining at 8-14 µSv—far below CBCT's 10-130 µSv range.^[60] Clinically, dental tomosynthesis supports endodontic, periodontal, and implant procedures by enabling better detection of hidden lesions, such as periapical abscesses or bone loss, and aids orthodontic planning through clearer root alignment views. Its advantages include dose reduction by up to 90% relative to CBCT for targeted diagnostics, faster acquisition without repositioning errors, and compatibility with existing workflows, as sIOT devices like the Portray system integrate seamlessly with digital sensors. Limitations persist in the form of incomplete volumetric data due to the restricted scan angle, potentially requiring supplementary imaging for deep structures, and while preclinical studies show superior pathology visibility, broader clinical validation has been pursued through trials such as the completed NCT02873585.^[59][60][61]

Industrial Applications

Tomosynthesis, particularly digital tomosynthesis (DT), has emerged as a valuable tool in industrial non-destructive testing (NDT) for inspecting complex structures without disassembly, offering three-dimensional imaging through limited-angle X-ray projections. This technique bridges the gap between two-dimensional radiography and full computed tomography (CT), enabling faster scans with fewer projections while providing depth-resolved images suitable for quality control in manufacturing processes.^[62][63] In additive manufacturing, tomosynthesis facilitates the inspection of large, thick components where traditional CT struggles due to penetration limitations and high costs. For instance, it has been applied to directed energy deposition-arc (DED-Arc) steel parts, such as a 106 mm × 106 mm × 27 mm flat wall, using a 320 kV X-ray source to acquire 100 projections over a 60° angle; this method reveals all defects visible in CT while outperforming conventional radiography by reducing artifacts from surface irregularities. The approach employs a fixed part and detector with a moving source, followed by deblurring reconstruction to generate slice images, cutting inspection time and enabling routine use in metal additive manufacturing R&D and production.^[63][62] For aerospace composites, such as glass fiber reinforced polymers (GFRP), DT integrated with deep learning enhances defect detection like delaminations and voids. Systems using low-power X-rays produce tomosynthetic slices that, when processed with models like Detectron2 (a Mask R-CNN variant), achieve a Dice coefficient of approximately 86% and intersection over union of 86%, enabling precise segmentation in near real-time for automated inspections. This combination reduces manual errors and supports integration with robotic systems, addressing challenges in high-stakes aerospace quality assurance.^[64] Tomosynthesis also supports battery manufacturing by allowing rapid 3D X-ray evaluations during R&D and assembly, identifying internal anomalies without extensive infrastructure. In welding applications, portable DT systems enable on-site 3D inspections for maintenance, improving defect localization in joints and accelerating workflow in industrial settings. Additionally, feasibility studies have demonstrated its use for multilayer printed circuit boards, reconstructing clear in-plane images with just 11 projections over a 20° scan angle via filtered backprojection in cone-beam geometry, highlighting its efficiency for electronics NDT. Real-time DT variants further extend to specialized inspections, such as solid rocket motors, where enhanced radioscopic differentiation aids in detecting embedded flaws.^{[62][62][65][66]} Overall, these industrial implementations leverage tomosynthesis's benchtop compatibility and reduced scan times—often compressing cycles by up to 65% in preform inspections—to lower costs, minimize waste, and enhance risk management across sectors like composites, energy storage, and propulsion.^[62]

Advantages and Limitations

Key Benefits

Tomosynthesis provides three-dimensional imaging capabilities with limited-angle projections, significantly reducing tissue superposition compared to conventional two-dimensional radiography, which enhances the visibility of anatomical structures and lesions obscured by overlapping tissues.^[10] This reduction in anatomic noise improves diagnostic accuracy across various applications, such as detecting pulmonary nodules in chest imaging, where studies have shown tomosynthesis to be up to three times more effective than standard radiography.^[67] In musculoskeletal evaluations, it offers superior depiction of bone fractures and joint pathologies by minimizing superimposition artifacts, allowing for better assessment of healing progress post-fixation.^[56] A primary advantage of tomosynthesis is its lower radiation dose relative to computed tomography (CT), often delivering 88% less exposure while providing pseudo-three-dimensional views suitable for routine screening and follow-up.^[68] For instance, in chest imaging for emphysema patients, it enables detailed visualization of lung parenchyma with reduced patient risk and financial burden compared to CT.^[69] In breast imaging, digital breast tomosynthesis (DBT) increases invasive cancer detection rates by 1.2 to 2.7 per 1,000 women screened and reduces false-positive recalls by 15% to 40%, particularly benefiting those with dense breasts.^[70] Similarly, in dental applications, intraoral tomosynthesis improves lesion characterization with decreased radiation and enhanced three-dimensional information over traditional intraoral radiography.^[59] Tomosynthesis also facilitates faster image acquisition and lower operational costs than full CT scans, making it a practical alternative for resource-limited settings or serial monitoring.^[71] In industrial contexts, cone-beam tomosynthesis extracts internal cross-sectional views efficiently, aiding in flaw detection like voids and cracks in materials without the need for extensive rotational scans.^[72] Overall, these benefits stem from its ability to balance enhanced depth resolution with accessibility, positioning tomosynthesis as a valuable tool in both clinical and non-clinical imaging.^[40]

Technical Challenges

One of the primary technical challenges in tomosynthesis is the limited angular range of x-ray projections, typically spanning 15° to 50°, which results in incomplete data sampling compared to full computed tomography (CT). This limited-angle acquisition leads to out-of-plane artifacts, where high-contrast structures appear as blurred or ghosted replicas in adjacent slices, compromising depth resolution and lesion localization.^[26] In breast tomosynthesis, for instance, these artifacts can obscure microcalcifications or masses, necessitating advanced filtering techniques like slice thickness filters to mitigate blurring.^[26] Image reconstruction poses significant computational demands due to the ill-posed nature of inverting limited projections. Traditional filtered backprojection (FBP) methods suffer from noise amplification and artifacts, while iterative algorithms such as maximum likelihood expectation maximization (MLEM) or simultaneous algebraic reconstruction technique (SART) offer better noise suppression but require substantial processing time—often minutes per volume without hardware acceleration like GPUs.^[26] In chest tomosynthesis, where respiratory motion exacerbates blurring over 10–15 second scans, these methods must incorporate motion compensation to maintain diagnostic utility, though depth resolution remains inferior to CT at approximately 5–10 mm slice thickness.^[73] Radiation dose management is another critical hurdle, as tomosynthesis acquisitions involve multiple low-dose projections that can cumulatively approach or exceed standard radiography levels—e.g., mean glandular dose in breast imaging can increase by 50–120% over 2D mammography for combined exams, depending on breast thickness, glandular fraction, and system geometry.^[10][74] However, the use of synthetic 2D mammography reconstructed from tomosynthesis projections can reduce the total dose to levels comparable to 2D alone.^[75] Scatter radiation further degrades contrast by up to 30% without anti-scatter grids, which are often omitted to minimize dose per projection, requiring software-based corrections that add complexity.^[10] Patient motion artifacts are particularly pronounced in non-breast applications like chest or dental imaging, where involuntary movements (breathing or swallowing) during extended scan times can introduce distortions not easily correctable in real-time.^[73] In dental tomosynthesis, limited field-of-view and jaw positioning challenges amplify these issues, demanding faster acquisition protocols or stationary source arrays to reduce exposure time below 5 seconds. Overall, these challenges drive ongoing research into hybrid reconstruction models and dose-efficient detectors to balance image quality with clinical feasibility.^[26]