Chapter Four · failure evidence
What Magnetic Resonance Imaging Techniques got wrong, from 36 dissertations
The reviewed records detail technical challenges and methodological failures encountered across magnetic resonance imaging and spectroscopy investigations. These difficulties encompass hardware and gradient encoding limitations, acquisition duration constraints, field inhomogeneity artifacts, low-field signal losses, and quantitative modeling inaccuracies. These records come from PhD theses at 19 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Machine learning and statistical processing pipelines failed to generalize or degraded image information
High-dimensional feature concatenation, unrepresentative retrospective training, and multi-slice convolutional architectures degraded segmentation and classification accuracy. Similarly, overly conservative statistical corrections, poorly chosen filtering parameters, and global clustering obscured regional patterns and produced high error rates.
Tried and failed
multimodal feature concatenation with neuroimaging applied to cognitive decline trajectory prediction. Outcome: worse than baseline. Reason: adding high-dimensional fMRI features degraded classifier performance compared to biomarker baselines
Developing machine learning and optimization models to classify Super-Agers and Cognitive Decliners · Iowa State
Considered and rejected
Considered and rejected: Rejected standard/global 1D clustering (e.g. k-means across all features/subjects) because it overgeneralizes and obscures localized, subpopulation-specific neuroimaging patterns
Granular factoring into neuroimaging dynamic across space, time, and modality · Georgia Tech
Tried and failed
retrospectively undersampled training for accelerated imaging reconstruction applied to prospective magnetic resonance image reconstruction. Outcome: did not generalise. Reason: prospective acquisition contains spatial distortions and contrast shifts not simulated by retrospective undersampling
Tried and failed
standard high-pass filtered phase processing applied to susceptibility-weighted magnetic resonance imaging. Outcome: worse than baseline. Reason: suboptimal echo time selection degraded lesion-to-vein contrast
Tried and failed
2.5D fully convolutional network applied to cardiac magnetic resonance volumetric image segmentation. Outcome: worse than baseline. Reason: Inter-slice spatial correlations provided less effective features than pure 2D slice-wise representation
Tried and failed
linear regression for binary-derived continuous score prediction applied to neuroimaging biomarker prediction. Outcome: worse than baseline. Reason: produced very low specificity and high false positive rates compared to logistic classification
Statistical and Machine Learning Methods for Neuroimaging and Neurocognitive Data · Penn
Tried and failed
mass-univariate voxelwise testing with Bonferroni correction applied to spatial functional imaging data. Reason: overly conservative thresholding led to high false negative rates (>30%)
Next-generation functional data analysis for imaging and spatial data · Iowa State
Non-standard gradient trajectories and novel spatial encoding mechanisms caused severe signal dephasing and hardware bottlenecks
Mechanical rotation suffered from slow speeds, excessive costs, and actuator precision limits compared to electronic coils. Furthermore, sinusoidal trajectories, super-resolution without motion, bandwidth mismatches in low-field gradients, and novel dipole coil geometries introduced dephasing artifacts and inferior signal-to-noise ratios.
Tried and failed
mechanical rotation for spatial encoding applied to magnetic resonance imaging. Outcome: too slow. Reason: slow speed, high cost, and actuator precision limitations compared to electronic gradient coils
Handheld MRI for Point-of-Care and Educational Applications · MIT
Tried and failed
introducing spatial magnetic field gradient for encoding applied to portable low-field magnetic resonance imaging. Outcome: worse than baseline. Reason: gradient bandwidth exceeded RF excitation bandwidth, causing non-uniform excitation and severely degrading SNR
Handheld MRI for Point-of-Care and Educational Applications · MIT
Tried and failed
software-only in-plane super-resolution without physical displacement applied to magnetic resonance imaging. Reason: scanner hardware constraints prevent field-of-view shifts along required axes without physical target motion
Image Guided High Precision Robotic Positioning in MRI for Medical Applications · Georgia Tech
Lost to a baseline
In small-sample finger imaging, peak sagittal SNR was higher for conventional LC loop (135.9 dB A.U.) and peak axial SNR for solenoid (129.9 dB A.U.) than solenoidal dipole (123 dB A.U.).
Advanced RF Hardware Solutions for Improved Imaging Performance and Safety in Ultra-High and Very-Low-Field MRI · DSpace at SUNY Buffalo
Tried and failed
sinusoidal gradient encoding with temporal shuffling applied to fast 3D magnetic resonance imaging. Reason: non-trivial intra-voxel dephasing caused severe image artifacts
On Improving the Acquisition and Reconstruction Of Spatio-Temporal Magnetic Resonance Imaging · MIT
Magnetic field inhomogeneities and flow phenomena generated severe geometric and reconstruction artifacts
Susceptibility differences around tissue boundaries and off-resonance effects caused severe image artifacts that prevented quantitative cardiac and spinal evaluations. Complex blood flow separation distorted vessel lumen estimates, while aggressive optimization penalties and banding handling amplified reconstruction noise and shadow artifacts.
Considered and rejected
Considered and rejected: Gradient-echo based functional imaging was rejected in favor of spin-echo based methods (HASTE) for spinal cord imaging due to magnetic field inhomogeneities caused by surrounding vertebrae and tissue interfaces.
Functional MRI Investigation of the Neural Basis of Human Pain Processing in the Brainstem and Spinal Cord; Descending Regulation and Chronic Pain · Queens University Institutional Repository
Tried and failed
T2* magnetic resonance relaxometry mapping applied to in vivo cardiac tissue characterisation. Reason: poor image quality and severe off-resonance artefacts prevented quantitative analysis
Tried and failed
time-of-flight magnetic resonance angiography applied to luminal defect volume estimation. Outcome: worse than baseline. Reason: flow artifacts and complex flow separation caused severe overestimation of tissue geometry compared to standard anatomical imaging
4D Flow Magnetic Resonance Imaging and Computational Fluid Dynamics for the Study of Complex Blood Flow Patterns in Carotid Webs · Georgia Tech
Tried and failed
unrolled ADMM optimization with high penalty parameter applied to magnetic resonance quantitative susceptibility mapping. Outcome: unstable. Reason: excessive penalty parameter weighting caused divergent fidelity costs and shadow artifacts around high-contrast regions
Lost to a baseline
In phantom regions with smooth, low-intensity tissue relaxation, an 8-phase-cycle discrete acquisition (DPC-8) achieved a lower coefficient of variation (CoV) than fm-bSSFP min-align due to reconstruction noise artifacts dominating imperceptible bands.
Debanding in Frequency-Modulated bSSFP MRI: High-Resolution Neuroimaging · JScholarship
Mathematical modeling approximations and correction schemes compromised quantitative relaxation and diffusion parameter estimation
Inaccurate deformation assumptions and unstable multi-compartment formulations caused systematic deviations and parameter estimation failures in diffusion and relaxometry imaging. In addition, polynomial calibrations failed across heterogeneous tissues, and correction adjustments for echo times and relaxation exacerbated noise and distorted fraction estimates.
Tried and failed
strain correction model with simplified deformation assumptions applied to diffusion tensor magnetic resonance imaging. Outcome: did not generalise. Reason: oversimplified deformation assumptions caused deviations from measured experimental diffusion data
Tried and failed
gradient-echo shifting in multi-echo sequence applied to quantitative T2 relaxometry myelin water imaging. Outcome: worse than baseline. Reason: increased minimum echo time caused systematic overestimation of short T2 component fraction
Quantitative Microstructural Imaging for Clinical Use · EPFL
Considered and rejected
Considered and rejected: Multi-echo T2 and mcDESPOT relaxometry were rejected for myelin water imaging due to model estimation instability and failure to account for compartmental exchange
Physiology and plasticity of myelinated long-range projections in the adult brain · Oxford
Tried and failed
polynomial fitting on coarse tissue segmentations applied to inhomogeneity calibration in magnetic resonance imaging. Outcome: did not generalise. Reason: failed to accurately model low-amplitude signal regions across heterogeneous tissue types
Imaging Glutamate In The Human Brain At Ultra-High Magnetic Field: Advances And Applications · Penn
Tried and failed
spatially-varying T2* correction applied to diffusion tensor magnetic resonance imaging. Outcome: worse than baseline. Reason: caused significant noise enhancement, reducing accuracy and precision of derived diffusion parameters
Prolonged acquisition times and rigid patient positioning restricted clinical feasibility for dynamic and functional imaging
Extended scan times requiring several hours or exceeding single breath-hold durations prevented the practical deployment of diffusion and angiographic protocols. Additionally, standard supine cross-sectional setups failed to represent anatomically relevant functional configurations found in upright postures.
Considered and rejected
Considered and rejected: Supine cross-sectional imaging (CT and MR angiography) rejected as a primary dynamic diagnostic modality because supine positioning fails to reflect the functional thoracic outlet anatomy present in upright postures
Arterielles Thoracic Outlet Syndrome: Die Bedeutung der Farbkodierten Duplexsonografie in der vaskulären Diagnostik Arterial thoracic outlet syndrome: the importance of color-coded duplex sonography in vascular diagnostics · open_UMR Marburg DSpace 10.0
Considered and rejected
Considered and rejected: Rejected PC-MRA and 3D TOF-MRA for abdominal vascular imaging due to excessive scan acquisition times incompatible with single breath-holds.
MRI changes in visceral fat in Crohn’s Disease · University of Nottingham Repository
Considered and rejected
Considered and rejected: Diffusion tensor magnetic resonance imaging (dt-MRI) rejected due to acquisition requiring several hours making it non-viable for dynamic imaging
Rapid Polarized Spatial Frequency Domain Imaging for Mapping Dynamic Fiber Orientation in Heart Valves · UT Austin
Low magnetic field strength and unshielded magnetometry suffered from prohibitive signal loss and spectral artifacts
Operating at low magnetic field strengths caused split-peak fitting artifacts in biofluid diffusion dimensions and severe signal loss during rotational correction. Coherence magnetometry protocols similarly proved unviable due to weak signal amplitudes and extreme sensitivity to background gradients.
Tried and failed
Intensity attenuation correction at large rotation angles applied to Multi-orientation low-field magnetic resonance imaging. Outcome: no signal. Reason: Severe SNR loss beyond thirty degrees of rotation degraded image quality substantially
In vivo imaging of collagen fibre structures using dedicated low-field MRI · Imperial
Tried and failed
diffusion-ordered NMR spectroscopy at low field strength applied to complex biofluid mixtures. Reason: lower magnetic field strength caused split-peak fitting artefacts in the diffusion dimension
Considered and rejected
Considered and rejected: Rejected using coherence magnetometry (MGFID) directly for diffusion measurements due to low signal amplitude and extreme sensitivity to residual magnetic field gradients.
Left open by the authors
Problems the authors named and did not get to.
Left open
Determine whether weak fMRI functional dissociations across individuals stem from biological variation or imaging resolution limits. Blocker: Broad exploratory research question without a specified methodology or dataset
Examining the Organization and Functions of Parallel Association Networks within Individuals · Harvard
Left open
Test crossmodal compositional decoding in lmSTC using fMRI experiments with single sentence and image presentations containing delay periods. Blocker: Requires conducting new fMRI neuroimaging experiments with human subjects and specialized physical apparatus (MRI scanner).
The Neural Representation of Structured Complex Ideas in Different Formats · Harvard
Left open
Evaluate cross-participant fMRI semantic decoders on brain imaging data collected from individuals with brain lesions and language disorders like aphasia. Blocker: Requires fMRI acquisition apparatus, clinical access to patients with aphasia/brain lesions, and specialized human subjects data.
Modeling and decoding semantic representations in language and vision · UT Austin
Left open
Analyze fMRI BOLD activation outside the visual cortex to study cognitive modulation during geospatial mental rotation tasks. Blocker: Requires private 3T fMRI experimental neuroimaging dataset and participant behavioral data from the thesis study
Neurological Role of Cartographic Visual Contrast in Geospatial Cognition · Scholars' Bank
Left open
Adjudicate whether the supramarginal gyrus encodes phrasing-invariant semantic information, functional affordances, sequence order, or phonological representations using targeted fMRI MVPA experiments. Blocker: Requires collecting new human fMRI neuroimaging data or access to private trial-level fMRI datasets with specific linguistic controls.
The Neural Representation of Structured Complex Ideas in Different Formats · Harvard
Left open
Identify specific shared kinematic, physical, and semantic features encoded by common neural representations across actions and object events using fMRI. Blocker: Requires collecting new controlled fMRI experiment data or accessing proprietary neuroimaging datasets.
Agents, objects, and actions: Investigations into the neural representation of dynamic information · Harvard
Left open
Apply the pervasive bias eIRT model to multi-test cognitive battery datasets and functional neuroimaging data. Blocker: Access to specific multi-test cognitive battery and paired in situ functional neuroimaging datasets with demographic variables.
Explaining Racial Discrepancies in Verbal Memory Assessment with Joint Estimation of Life Course Social Inequalities and Measurement Bias · DSpace at UTSWMED
Left open
Conduct in vivo studies to evaluate the sensitivity and reliability of higher-order radiomic parameters in diagnosing tendinopathy. Blocker: Requires in vivo clinical/animal studies, medical ultrasound imaging, and physical subjects.
Radiomic Analysis of Sonograms for the Detection of Tendon Damage · Virginia Tech
Left open
Optimize a 3T MRI sequence balancing spatial resolution and acquisition time for imaging glenoid cartilage wear in shoulder hemiarthroplasty. Blocker: Requires access to a 3T MRI scanner and physical shoulder specimens/patients.
Left open
Perform muscular segmentation and analysis on high-resolution 3T MRI VIBE sequence spinal and lower-limb scans. Blocker: Requires private 3T MRI patient imaging data collected during the study.
Understanding spinal loading in persons with traumatic lower limb amputations · Imperial
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.