MR Glossary

Turkish: Üç boyutlu görüntüleme

Imaging in which a whole volume is excited and encoded with phase encoding in two directions, instead of slice by slice. It gives thin, gap-free slices and high SNR, MPRAGE being an example.

Turkish: Akustik gürültü

Loud sound from vibrating gradient coils. It can exceed 100 dB, so hearing protection is used in every scan.

Turkish: Katlanma artefaktı

Tissue outside the field of view folding onto the opposite side of the image. It usually happens along the phase-encoding direction. A larger field of view or oversampling fixes it.

Turkish: Anomali tespiti

A model trained only on normal images flagging regions that depart from that norm. It is used to find lesions when labelled patient data are scarce.

Turkish: Görünür difüzyon katsayısı

A quantitative measure of how fast water diffuses in tissue, in mm²/s. In acute stroke diffusion is restricted and ADC drops.

Turkish: Arteriyel spin etiketleme

A perfusion method that magnetically labels water in arterial blood and measures it once it reaches tissue. It quantifies cerebral blood flow without a contrast agent.

Turkish: Yapay zeka

The broad name for computer methods that learn patterns from data to make decisions or produce outputs. In MRI it is used to speed up scans, improve image quality, segment structures and support diagnosis.

Turkish: Atlas ve parselasyon

A ready-made map that divides the brain into anatomical or functional regions. It lets data in a common space be summarised by region and connectivity networks be built.

Turkish: Dikkat mekanizması

A way for a network to learn which parts of the input matter most for each output and weight them accordingly. Its cost grows with the square of the input length, so it can be expensive for high-resolution images.

Turkish: b değeri

The strength of diffusion weighting, in s/mm². It depends on gradient strength, duration and spacing. A b = 0 image has no diffusion weighting. Around 1000 s/mm² is common in brain imaging.

A set of rules for organising brain imaging data with standard file names and folders (Brain Imaging Data Structure). It makes data easier to share and to process with automated pipelines.

Turkish: Blok tasarım

An experimental layout in which conditions alternate in blocks, typically of 15–30 seconds. It has high statistical power.

Turkish: Kan oksijenlenme düzeyine bağlı sinyal

Deoxyhaemoglobin is paramagnetic and lowers T2* signal. Blood flow to an active region rises more than oxygen use, so the deoxyhaemoglobin fraction falls and the signal rises by a few percent.

Turkish: Beyin yaşı tahmini

A model trained on MRI scans of healthy people predicting the apparent age of a new person's brain. The gap between predicted and chronological age is studied as a marker of ageing and disease.

Turkish: Kardiyak ve solunum tetikleme

Synchronising data acquisition with the heartbeat or breathing, using ECG, a pulse sensor, a respiratory belt or navigator signals. It reduces motion artifacts.

Turkish: Kimyasal kayma

The slightly different resonance frequency of the same nucleus in different molecules. Water and fat protons differ by about 3.4 ppm, roughly 440 Hz at 3 T.

Turkish: Kimyasal kayma artefaktı

Fat signal shifting along the readout direction because of the fat-water frequency difference. Bright and dark bands appear at tissue boundaries. Raising the bandwidth reduces it.

Turkish: Sınıf dengesizliği

One class having far fewer examples than the others, as with a rare disease. Accuracy becomes misleading, and weighting, resampling and balanced metrics are needed.

Turkish: Sınıflandırma

A task that assigns an image or a person to one of several predefined classes, such as patient versus healthy control. Performance is measured with accuracy, sensitivity, specificity and AUC.

Turkish: Küme tabanlı çıkarım

Applying a voxel threshold first and then declaring large enough clusters of adjacent voxels significant. A lenient initial threshold inflates the false-positive rate.

Turkish: Bobin birleştirme

Merging the images from each channel of a multichannel receiver into one image. The simplest method is the root sum of squares (RSS). Combination with coil sensitivity maps preserves phase.

Turkish: Sıkıştırılmış algılama

A method that exploits the sparsity of the image in a transform domain to reconstruct, iteratively, from incoherently undersampled data.

Turkish: Kısıtlı küresel ters evrişim

A method that deconvolves the diffusion signal with the response of a single fibre bundle to estimate the fibre orientation distribution (FOD) in each voxel. It can resolve crossing fibres and is a basis for tractography.

Turkish: Kontrast-gürültü oranı

The signal difference between two tissues divided by the noise. It measures how well a structure stands out from its surroundings.

Turkish: Evrişimli sinir ağı

A neural network that learns local features with small filters sliding over the image. It has long been the standard architecture for image tasks such as denoising, reconstruction and segmentation.

Turkish: Çapraz doğrulama

Splitting data into folds and repeatedly training and testing the model, holding out a different fold each time. If model settings are also being tuned, nested cross-validation should be used.

Turkish: Kesişen lifler

More than one fibre bundle running in different directions within a voxel. It occurs in most white-matter voxels and is the main limitation of the single-direction tensor model.

Turkish: Veri artırma

Varying training images with rotations, shifts, contrast changes and similar operations so the model learns more general features.

Turkish: Veri tutarlılığı

A step that keeps the network's output consistent with the k-space data that were actually measured. Measured points are kept, and the network only fills in what is missing. It reduces the risk of hallucination.

Turkish: Veri sızıntısı

Information from the test data entering the training process, for example selecting features on the whole dataset or having the same person's scans in both training and test sets. It makes performance look far better than it is.

Turkish: Derin öğrenme

A machine learning approach that learns directly from data with multi-layer neural networks, without hand-crafted features. Most current MRI applications build on it.

Turkish: Derin öğrenme tabanlı geriçatım

Reconstruction that uses trained neural networks to recover images from undersampled or noisy data. Methods that build the physics model into the network tend to be more reliable. Results should be checked for structures the network may hallucinate.

Turkish: Gürültü giderme

Reducing random noise in an image while keeping detail. Learning-based denoising is used to make up for the SNR lost in short or high-resolution scans.

Turkish: Tanısal duyarlılık ve özgüllük

Sensitivity is the share of truly affected cases that are correctly detected, specificity the share of unaffected cases that are correctly ruled out. Not to be confused with magnetic susceptibility in MRI.

Turkish: Dice katsayısı

A measure of overlap between two segmentations, from 0 to 1, where 1 means perfect overlap. It is used to compare automatic segmentation with expert outlines.

Turkish: Difüzyon verisinde gürültü giderme

Reducing noise by exploiting the shared structure across many diffusion images. Marchenko-Pastur principal component analysis (MP-PCA) is a common method and is usually the first preprocessing step.

Turkish: Difüzyon yönleri (b vektörleri)

The directions along which diffusion gradients are applied. Each diffusion-weighted image corresponds to one direction. A tensor needs at least 6, and many more are used to resolve crossing fibres. Directions must be rotated when images are registered.

Turkish: Difüzyon kurtosis görüntüleme

An extension of the tensor model that measures how far water diffusion departs from a Gaussian distribution. It is sensitive to microstructural complexity and needs at least two non-zero b-values.

Turkish: Difüzyon modeli

A generative model that learns to reverse a process that adds noise to an image step by step. It produces an image by starting from pure noise and removing it gradually. Quality is high, but the many steps can make it slow. It is unrelated to diffusion-weighted MRI.

Turkish: Difüzyon tensör görüntüleme

A method that measures diffusion along many gradient directions and estimates a tensor in each voxel. Fractional anisotropy (FA) maps and white-matter tractography are derived from it.

Turkish: Difüzyon ağırlıklı görüntüleme

Imaging that is sensitive to the random motion of water molecules. Strong gradients reduce the signal of moving spins. Sensitivity is set by the b-value (s/mm²), and results are also shown as apparent diffusion coefficient (ADC) maps.

Turkish: Dixon yöntemi

A method that uses the phase difference between water and fat to compute separate water and fat images from images taken at a few echo times. It gives robust fat suppression and quantifies fat fraction.

Turkish: Alan uyarlaması

Methods that adapt a model trained in one domain to another, with few examples or unlabelled data. They matter for models that must work on data from different sites.

Turkish: Dağılım kayması

The data a model meets differing from its training data, for example another scanner, field strength, sequence or patient group. Performance can drop sharply.

Turkish: Eko

A regrowth of decayed signal at a set time, created by restoring phase coherence. Refocusing uses a 180° RF pulse (spin echo) or a reversal of the gradient (gradient echo).

Turkish: Ekoplanar görüntüleme

A sequence that fills all or most of k-space rapidly after a single excitation, imaging a slice in well under a second. It underpins fMRI and diffusion imaging but is prone to distortion.

Turkish: Eko zamanı

The time from excitation to the centre of the echo. A long TE emphasises T2 (or T2*) differences.

Turkish: Girdap akımı düzeltmesi

Removing direction-dependent image distortions caused by eddy currents that strong diffusion gradients induce in conductive parts. It is usually done together with head motion correction.

Turkish: Kodlama modeli

A model that predicts each voxel's response from features of the stimulus or task. With rich stimuli such as natural films or speech it maps which information the brain represents where. It is the reverse of decoding.

Turkish: EPI bozulması

Stretching or compression of EPI images caused by field inhomogeneity and the low bandwidth along the phase-encoding direction. It can be corrected with an extra scan using reversed phase encoding.

Turkish: Olay ilişkili tasarım

An experimental layout in which brief stimuli are presented one at a time, often with jittered intervals. It lets responses to individual events be separated.

Turkish: Açıklanabilirlik

Showing which image regions or features an AI model based its decision on. It matters for trust and debugging in clinical use.

Turkish: Yağ baskılama

Techniques that reduce fat signal so nearby tissues stand out. Frequency-selective saturation, STIR and Dixon methods are the most common.

Turkish: Öznitelik çıkarımı

Deriving numerical descriptors from images for a model to use. Regional volumes, cortical thickness and connectivity values are hand-crafted features. Deep networks learn features directly from data.

Turkish: Federe öğrenme

Several institutions training a shared model without exchanging patient data. Each site updates the model on its own data, and only the updates are combined. It allows learning from more varied data while protecting privacy.

Turkish: Alan haritası

An extra scan that measures main-field inhomogeneity voxel by voxel. It is used to correct distortion in EPI images. It can also be estimated from a pair of EPI images with reversed phase encoding.

Turkish: Görüş alanı

The physical size of the imaged region, usually in millimetres. If the object is larger than the field of view, aliasing can occur.

Turkish: İnce ayar

Adapting a pretrained model to a new task or dataset with further training. Parameter-efficient methods such as LoRA train only a small set of added parameters, cutting compute and memory needs.

Turkish: Çevirme açısı

How far an RF pulse tips the net magnetization away from B₀. A 90° pulse moves it fully into the transverse plane. Gradient echo sequences usually use smaller angles.

Turkish: Sıvı baskılı evirme geri kazanımı

A T2-weighted sequence whose inversion time nulls the cerebrospinal fluid signal. Lesions next to the ventricles become easier to see.

Turkish: İleri model

The mathematical model of the physical process from image to measured data. In MRI it consists of coil sensitivities, the Fourier transform and the sampling mask. Physics-guided networks and data consistency rely on it.

Turkish: Temel model

A large model pretrained on vast and varied data that can be adapted to many tasks with little data. Such models are now being developed for medical imaging too.

Turkish: Fourier dönüşümü

A mathematical operation that splits a signal into its frequency components. In MR the inverse Fourier transform turns k-space data into an image.

Turkish: Fraksiyonel anizotropi

A measure from 0 to 1 of how strongly diffusion favours one direction. It is high in coherent white-matter bundles. It can be low where fibres cross, even in healthy tissue.

Turkish: Çerçeve yer değiştirmesi

A single-number summary of head motion between consecutive volumes, in millimetres. Volumes above a threshold can be removed from analysis (scrubbing) or flagged in the model.

Turkish: Serbest indüksiyon sönümü

The decaying signal picked up by the receive coil right after excitation. With no refocusing it decays with T2*.

Turkish: Frekans kodlama

Keeping a gradient on while the signal is read so that position maps to frequency. This direction is also called the readout direction.

Turkish: Saçak alan

The magnetic field that extends outside the magnet. People with active implants such as pacemakers are kept outside the 0.5 mT (5 gauss) line.

Turkish: Fonksiyonel bağlantısallık

The temporal relation between signals from different brain regions, usually measured as correlation. It shows regions working together, not direct anatomical links.

Turkish: Fonksiyonel MRG

MR imaging that tracks activity-related changes in blood oxygenation over time. It mostly uses BOLD contrast and EPI.

Turkish: Gadolinyum bazlı kontrast madde

An intravenous contrast agent that shortens T1 and makes the tissue it reaches bright on T1-weighted images. Its use needs care in people with poor kidney function and in pregnancy.

Turkish: Genel doğrusal model

A statistical framework that explains each voxel's time series as a linear sum of the task conditions convolved with the HRF. Activation maps come from the model's coefficients.

Turkish: Çekişmeli üretici ağ

A generative model made of two networks trained against each other: a generator that tries to make realistic images and a discriminator that tries to tell real from generated. It gives sharp images, but training can be unstable.

Turkish: Üretken model

A model that learns the probability distribution of data and can draw new samples from it. In MRI it is used to synthesise missing contrasts, recover images from undersampled data and augment training sets.

Turkish: Üretken önselli geriçatım

Guiding a generative model that has learned the image distribution, usually a diffusion model, to produce images consistent with the measured data. It adapts flexibly to acquisition settings, but can be slow and needs care against hallucination.

Turkish: Gibbs halkalanması

Fine parallel lines next to sharp signal edges. They come from truncating k-space. A larger matrix or filtering reduces them.

Turkish: Gradyan

An added field that changes magnetic field strength linearly along one direction. It encodes where the signal comes from. Strength is given in mT/m and slew rate in T/m/s.

Turkish: Gradyan bobini

The set of coils that produces gradient fields along x, y and z. Rapid current switching causes the loud scanner noise and can cause peripheral nerve stimulation.

Turkish: Gradyan eko

A sequence that forms the echo by reversing a gradient instead of applying a 180° pulse. With a small flip angle and short TR it is fast. Its contrast is sensitive to T2*.

Turkish: Çizge sinir ağı

A neural network that works on graph data made of nodes and edges. It is used to make predictions from connectome data, treating brain regions as nodes and connectivity values as edges.

A parallel imaging method that fills missing k-space lines from neighbouring lines, using weights learned from a fully sampled calibration region.

Turkish: Grup analizi

Combining individual contrast maps in a second-level model. Mixed-effects models that account for between-subject variability let results generalise to the population.

Turkish: Jiromanyetik oran

The constant linking a nucleus's resonance frequency to field strength. For hydrogen, γ/2π is about 42.58 MHz/T. Each nucleus has its own value, so the scanner is tuned to a specific nucleus.

Turkish: Hemodinamik yanıt fonksiyonu

The time course of the BOLD signal after a brief neural event. It peaks at about 5–6 seconds, then returns to baseline after a brief undershoot.

Turkish: Halüsinasyon

An AI model producing a realistic-looking structure that is not in the real data, or removing one that is. In medical imaging, a small lesion appearing or disappearing can have serious consequences.

Turkish: Harmonizasyon

Reducing systematic differences between data from different scanners and sites while keeping biological differences. Statistical methods such as ComBat or deep learning image translation are used.

Turkish: Yüksek açısal çözünürlüklü difüzyon

Acquiring data with many diffusion directions, often 60 or more. It makes it possible to resolve more than one fibre direction in a voxel.

Turkish: Yüksek geçiren filtre

Removing slow drifts in the fMRI time series caused by scanner heating and physiology. The cutoff is chosen so it does not remove the slowest change in the experimental design.

Turkish: Görüntü geriçatımı

Computing the image from the measured raw data. For fully sampled Cartesian data this is an inverse Fourier transform. Undersampled data need extra information or models.

Turkish: Görüntü çakıştırma

Estimating the transform that brings two images into spatial alignment. It can be rigid (6 degrees of freedom), affine (12) or nonlinear. Aligning a functional image to the same person's anatomical image is called coregistration.

Turkish: Görüntü sentezi

Producing an image of a contrast or modality that was not acquired, from the images that were. For example, FLAIR can be synthesised from T1- and T2-weighted images, or a CT-like image from MRI. It can shorten exams, but results must be validated before diagnostic use.

Turkish: Görüntüden görüntüye çeviri

A learning task that maps an image to another domain. It can be trained with paired data (two contrasts from the same person) or unpaired data. Image synthesis and denoising are examples.

Turkish: Örtük sinirsel temsil

Representing an image not as a pixel array but as a small neural network that takes a coordinate and returns the intensity there. It can be used for reconstruction from a single scan's data, with no training set.

Turkish: Bağımsız bileşen analizi

A method that splits data into statistically independent spatial or temporal components. In fMRI it is used to find resting-state networks and to identify components caused by motion or physiological noise.

Turkish: Ters problem

The problem of recovering the unknown object that produced the measured data. MR reconstruction is an inverse problem that recovers an image from incomplete, noisy k-space data and usually has no unique solution without prior information.

Turkish: Evirme geri kazanımı

A preparation that applies a 180° inversion pulse and waits for magnetization to recover. It is used to boost T1 contrast or to null a chosen tissue.

Turkish: Evirme zamanı

The time between the inversion pulse and the excitation pulse. Exciting when a tissue's signal crosses zero suppresses that tissue.

Turkish: k-uzayı

The spatial-frequency domain where raw data are stored. The image is obtained by a Fourier transform of k-space. The centre carries contrast, the edges carry fine detail.

Turkish: Larmor frekansı

The frequency at which spins precess around the magnetic field. It is proportional to field strength. For hydrogen it is about 64 MHz at 1.5 T and about 128 MHz at 3 T.

Turkish: Sıvı helyum

The coolant that keeps superconducting windings at operating temperature. It boils at about 4.2 K. Closed cooling systems keep losses to a minimum.

Turkish: Manyetik parçacık görüntüleme

A modality distinct from MRI that directly measures the magnetic response of superparamagnetic iron oxide nanoparticles. A field-free point or line is scanned to map where they are and how concentrated. There is no tissue background.

Turkish: Manyetik rezonans

Nuclei placed in a strong magnetic field absorb and re-emit radio waves at their resonance frequency. MR imaging measures this signal, mostly from hydrogen nuclei in the water and fat of tissue.

Turkish: Manyetizasyon transferi

The transfer of saturation from protons bound to large molecules to the free water signal. It gives contrast sensitive to macromolecular content such as myelin.

Turkish: Ana manyetik alan

The scanner's strong, static magnetic field, which is always on. Its strength is given in tesla (T). UMRAM runs 3 T and 1.5 T research scanners.

Turkish: Matris

The number of pixels in each direction of the image, for example 256 × 256. A larger matrix at the same field of view gives finer detail but lowers SNR and can lengthen the scan.

Turkish: Ortalama, eksenel ve radyal difüzivite

Measures derived from the tensor. MD is the average diffusion over all directions, AD the diffusion along the main fibre direction and RD across it. They can help tell myelin changes from axonal changes.

Turkish: Hareket artefaktı

Blurring and ghosting from patient motion, breathing, heartbeat or blood flow during the scan. Ghosts usually appear along the phase-encoding direction.

Turkish: Hareket düzeltme

Aligning the images of a time series to remove head motion. Even a few millimetres of motion can create spurious activation or connectivity.

Magnetization-prepared rapid gradient echo: a high-resolution 3D T1-weighted sequence prepared with an inversion pulse. It is the usual anatomical reference scan in brain research.

Turkish: MR güvenli, MR koşullu, MR güvensiz

Standard labels for implants and devices. MR Safe poses no risk in any MR environment. MR Conditional may be scanned only under the conditions the manufacturer specifies, such as field strength and SAR. MR Unsafe must never enter the magnet room.

Turkish: MR güvenlik bölgeleri

The division of an MR unit into four zones by access level. Zone IV is the magnet room, entered only by screened people accompanied by trained staff.

Turkish: MR spektroskopi

A method that measures the spectrum of chemical compounds in a volume instead of an image. In the brain, metabolites such as NAA, choline, creatine and lactate are told apart by their chemical shifts.

Turkish: Gebelikte MR

Whether a pregnant patient has a clinical MRI is a physician's decision, and gadolinium-based contrast is generally avoided. In research scans, pregnancy is usually an exclusion criterion and is asked on the screening form.

Turkish: Çok kabuklu difüzyon

Acquiring data at more than one b-value, each with many directions. Models beyond the tensor (NODDI, DKI) and methods that separate tissue types need such data.

Turkish: Çok kipli öğrenme

Learning that combines different MR contrasts, other imaging modalities, clinical data or text. The sources complement each other. Coping with missing modalities is an active research topic.

Turkish: Çoklu karşılaştırma düzeltmesi

Adjusting the threshold to limit chance findings when tens of thousands of voxels are tested separately. Either the family-wise error rate (FWE) or the false discovery rate (FDR) is controlled. Cluster-based and permutation tests are also common.

Turkish: Çok vokselli örüntü analizi

An analysis that decodes information, such as which stimulus was seen, from the joint activity pattern of many voxels instead of single voxels. Machine learning classifiers are often used.

Turkish: Net manyetizasyon

The summed magnetic moment of all spins in a volume. At equilibrium it points along B₀. The measured MR signal arises when this vector is tipped into the transverse plane.

Neurite orientation dispersion and density imaging: a biophysical model that estimates more tissue-specific measures, such as neurite density and orientation dispersion, from multi-shell data.

Turkish: Kartezyen olmayan örnekleme

Sampling k-space along radial spokes or spirals instead of straight lines. It can be more robust to motion or faster, but reconstruction is more complex.

Turkish: Normatif modelleme

Learning the normal distribution of a brain measure by age and sex from a large healthy sample, then measuring how far individuals deviate from it. It lets patients be studied as individuals rather than as group averages.

Turkish: Gürültü regresörleri

Variables of no interest added to the model so they can be removed from the signal, such as motion parameters, white matter and cerebrospinal fluid signals, or CompCor components. They matter especially in resting-state analysis.

Turkish: Aşırı öğrenme

A model memorising its training data and failing on new data. It is common when large models are trained on little data. Data augmentation and regularisation reduce the risk.

Turkish: Paralel görüntüleme

An acceleration method that skips part of k-space by using the different spatial sensitivities of multichannel receive coils. The speed-up is given as the factor R.

Turkish: Paralel iletim

Delivering the RF pulse through several independently controlled transmit channels. It is used at high field to even out B₁ and manage SAR. UMRAM's 3 T scanner supports a transmit array.

Turkish: Kısmi Fourier

An acceleration method that uses k-space symmetry to measure only a little over half of it. Scan time drops, with a small loss of SNR.

Turkish: Kısmi hacim etkisi

A voxel containing more than one tissue, so its signal is a mixture. It blurs the edges of small structures and affects volume measurements.

Turkish: Periferik sinir uyarımı

Tingling or muscle twitching caused by currents that rapid gradient switching induces in tissue. The scanner keeps gradient switching below this threshold.

Turkish: Faz kodlama

Applying a brief gradient before readout so that position is stored in the spin phase. Each phase-encoding step fills one line of k-space, and the number of steps largely sets scan time.

Turkish: Fizik güdümlü ağ

A deep learning method that builds the MR measurement model, the encoding and coil sensitivities, into the network. The steps of an iterative reconstruction algorithm are unrolled into network layers. It is more reliable and generalises better than purely data-driven networks.

Turkish: Fizyolojik gürültü

Fluctuations in the BOLD signal caused by breathing and heartbeat. They can be recorded with a respiratory belt and pulse sensor and removed with methods such as RETROICOR.

Turkish: Tak-çalıştır önseller

A method that replaces the regularisation step of an iterative reconstruction with a pretrained denoising network. The same network can be used with different sampling patterns without retraining.

Turkish: Presesyon

The rotation of a spin's axis around the field direction, like the axis of a spinning top. Its rate is the Larmor frequency.

Turkish: Ön işleme

The chain of steps that prepares raw fMRI or diffusion data for statistical analysis. For fMRI a typical order is motion correction, slice timing correction, distortion correction, registration to the anatomical image, normalisation and smoothing.

Turkish: Mahremiyet koruyan öğrenme

Methods that make it hard to recover personal data from a model or from shared information, during or after training. Federated learning, differential privacy and de-identification are among them.

Turkish: Fırlatma etkisi

Ferromagnetic objects being pulled into the magnet with great force. Scissors, oxygen cylinders or chairs can cause fatal injury. Because the magnet is always on, the risk exists outside scans too.

Turkish: Proton yoğunluğu

The amount of signal-producing hydrogen nuclei in a tissue. Proton density weighted images use a long TR and short TE to reduce T1 and T2 effects and bring out this difference.

Turkish: PSNR ve SSIM

Two common metrics for how closely a generated image matches a reference. PSNR looks at pixel differences, SSIM at structural similarity. High values do not always guarantee diagnostic value.

Turkish: Psikofizyolojik etkileşim

An analysis that tests how a region's connectivity with other regions changes with the task condition.

Turkish: Açık veri setleri

Openly shared image collections that let methods be compared. For example, fastMRI offers raw k-space data, IXI multi-contrast brain images and BraTS tumour segmentation data. Their terms of use and ethics approvals must be respected.

Turkish: Sayısal duyarlılık haritalama

A method that computes tissue magnetic susceptibility quantitatively from gradient echo phase data. It is used to study iron and myelin content.

Turkish: Söndürme

The rapid loss of superconductivity in the magnet. The helium boils off and is vented through the quench pipe. It can be triggered on purpose to shut down the field in an emergency. Gas leaking into the room creates a risk of asphyxiation.

Turkish: Radyomik

Extracting many texture, shape and intensity features from medical images and using them for diagnosis or outcome prediction. Because features depend on scanner and sequence settings, harmonisation and external validation are needed.

Turkish: Alıcı dizi bobin

A receiver built from many small coil elements, each with its own channel. It raises SNR and makes parallel imaging possible.

Turkish: Alıcı bant genişliği

The range of frequencies sampled during readout, often given in Hz/pixel. A higher bandwidth shortens readout and reduces chemical shift artifact, but lowers SNR.

Turkish: İlgi bölgesi

Restricting analysis to a predefined brain region. The region can come from an anatomical atlas, a separate localiser scan or earlier studies. It raises statistical power and reduces the multiple comparisons burden.

Turkish: Regresyon

Predicting a continuous value from images, such as age, a cognitive score or disease severity. Performance is measured with mean absolute error or correlation.

Turkish: Düzenlileştirme

A prior added to an inverse problem to pick a plausible solution among the many that fit the measurements. Total variation and wavelet sparsity are classic examples. In deep learning the prior is learned from data.

Turkish: Tekrarlama zamanı

The time between successive excitation pulses. A short TR emphasises T1 differences. Scan time is close to TR times the number of phase-encoding steps.

Turkish: Temsili benzerlik analizi

An analysis that summarises how similar the activity patterns for different stimuli are in a matrix and compares that structure with models or other regions.

Turkish: Dinlenim durumu fMRG

fMRI acquired while the participant performs no task. The synchrony of slow signal fluctuations between regions is studied as functional connectivity.

Turkish: Retinotopik haritalama

Mapping with fMRI how positions in the visual field are laid out across visual cortex. Rotating wedge and expanding ring stimuli or population receptive field (pRF) models are used.

Turkish: RF yanığı

A burn caused by RF energy concentrating at skin contact points, cables or conductive loops formed by the body. Keeping arms and legs from touching, keeping cables off the skin and using proper pads reduce the risk.

Turkish: RF bobini

The antenna that transmits RF pulses and receives the MR signal. The body coil is built into the bore. Dedicated coils placed close to the head, knee and other regions give higher SNR.

Turkish: RF manyetik alanı

The weak magnetic field produced during an RF pulse, perpendicular to B₀ and rotating at the Larmor frequency. At high field it may not spread evenly through the body.

Turkish: RF darbesi

A short burst of radiofrequency energy at the Larmor frequency. It tips the net magnetization away from equilibrium to create signal. Different pulses are used for excitation, refocusing and inversion.

Turkish: RF kafesi

A conductive enclosure around the magnet room that keeps outside radio signals from entering the image. If the door is not fully closed, interference lines can appear.

Turkish: ROC eğrisi ve AUC

A curve showing a classifier's sensitivity against false-positive rate across decision thresholds, and the area under it. An AUC of 0.5 means chance, 1 means perfect separation.

Turkish: Güvenlik tarama formu

A form completed before scanning that asks about implants, metal fragments, pregnancy and other risks. It is filled in for the participant and for anyone who will enter the magnet room with them.

Turkish: Bölütleme

Assigning every voxel to a structure or tissue, for example outlining brain tissues, a tumour or the hippocampus. It is a basic step for volume measurement and treatment planning.

Turkish: Öz-gözetimli öğrenme

An approach that derives a training signal from the structure of the data itself, with no labels or references. For example, measured k-space is split in two and the model learns to predict one part from the other. It is valuable when fully sampled data do not exist.

Sensitivity encoding: a parallel imaging method that unfolds aliased images in image space using coil sensitivity maps.

Turkish: Şim

Adjusting the main field to make it more uniform. Passive shimming uses pieces of metal, active shimming uses extra coils. Good shimming directly affects fat suppression, EPI and spectroscopy.

Turkish: Kısa TI evirme geri kazanımı

A sequence that suppresses fat signal using a short inversion time. It is less sensitive to field inhomogeneity than frequency-selective fat suppression.

Turkish: Sinyal-gürültü oranı

The ratio of image signal to noise. Field strength, voxel volume, coil, receiver bandwidth and number of averages all affect it. Quadrupling the averages doubles SNR.

Turkish: Eşzamanlı çok kesitli görüntüleme

Exciting several slices at once and separating them with receive coil sensitivities. In fMRI and diffusion imaging it gives whole-brain coverage at short TR.

Turkish: Kesit seçimi

Applying a gradient during the RF pulse so that only spins in one slice are excited. Slice thickness is set by the pulse bandwidth and the gradient strength.

Turkish: Kesit kalınlığı

The thickness of the excited slice in 2D imaging. Thin slices reduce partial volume effects but lower SNR.

Turkish: Dilim zamanlaması düzeltmesi

Removing the timing differences caused by slices of a volume being acquired at different moments, by shifting each slice's signal to a common time point. Its impact is smaller with short TR and simultaneous multi-slice imaging.

Turkish: Uzamsal normalizasyon

Transforming different people's brain images into a common template space, such as MNI space. It is a prerequisite for group analysis.

Turkish: Uzamsal yumuşatma

Blurring the image with a Gaussian kernel. The kernel width is given as full width at half maximum (FWHM), often 4–8 mm. It raises SNR and evens out anatomical differences across subjects, at the cost of spatial detail.

Turkish: Özgül soğurma oranı

The rate at which the body absorbs RF energy, in W/kg. The scanner monitors SAR continuously to limit tissue heating. In normal operating mode the whole-body limit is 2 W/kg.

The quantum-mechanical angular momentum of a nucleus. Because of its spin, the hydrogen nucleus (a proton) behaves like a tiny magnet and interacts with an external magnetic field.

Turkish: Spin eko

A sequence that forms an echo with a 180° refocusing pulse after a 90° excitation. It undoes field inhomogeneity effects, so it gives T2 contrast and is more robust to distortion from metal and air.

Turkish: Durum uzayı modeli

A network type that processes long sequences with compute that grows linearly with input length. Selective state-space models such as Mamba offer transformer-like long-range context at lower cost and are now used in medical image synthesis and reconstruction.

Turkish: İstatistiksel kontrast

A linear combination of GLM coefficients that expresses the question being tested, for example "faces minus houses". It yields a t or F statistic per voxel, shown as a statistical map.

Turkish: Yapısal bağlantısallık

Turning tractography pathways into a connection matrix between atlas regions. The brain's network structure is studied with graph-theory measures.

Turkish: Süper çözünürlük

Estimating a higher-resolution image from a lower-resolution one. It is used to obtain thin-slice images from thick-slice scans. Whether the estimated detail is real must be assessed carefully.

Turkish: Süperiletken mıknatıs

A magnet made of niobium-titanium windings cooled to about 4 kelvin by liquid helium. Current flows without resistance, so the field is always on, also when no scan is running.

Turkish: Gözetimli öğrenme

A model learning from examples where each input is paired with the correct output. For MR reconstruction this needs fully sampled reference images, which are not always available.

Turkish: Duyarlılık artefaktı

Signal loss and distortion where materials of different magnetic susceptibility meet, such as at air-tissue boundaries or around metal. It is pronounced in gradient echo and EPI.

Turkish: Duyarlılık bozulması düzeltmesi

Removing the stretching and compression that field inhomogeneity causes along the phase-encoding direction in EPI. It uses the field estimated from a pair of b = 0 images with reversed phase encoding, or from a field map.

Turkish: Duyarlılık ağırlıklı görüntüleme

A method that combines magnitude and phase of a gradient echo image to emphasise susceptibility differences. It is very sensitive to veins, microbleeds and iron deposits.

Turkish: T1 gevşemesi

The recovery of longitudinal magnetization after excitation (spin-lattice relaxation). T1 is the time for about 63 percent of that recovery. Fat has a short T1, cerebrospinal fluid a long one.

Turkish: T1 ağırlıklı görüntü

An image whose contrast comes mainly from T1 differences, made with a short TR and short TE. Fat appears bright and fluid dark, and it is widely used for anatomy.

Turkish: T2 gevşemesi

The decay of transverse magnetization as spins lose phase coherence through spin-spin interactions. T2 is the time for the signal to fall to about 37 percent, and in tissue it is always shorter than T1.

Turkish: T2 ağırlıklı görüntü

An image whose contrast comes mainly from T2 differences, made with a long TR and long TE. Fluid appears bright, so oedema and many lesions stand out.

Turkish: T2* gevşemesi

Faster signal loss that adds the effect of magnetic field inhomogeneity to T2 decay. It sets the contrast of gradient echo images and BOLD fMRI.

The SI unit of magnetic flux density. Clinical and research scanners mostly run between 1.5 T and 7 T. For comparison, the Earth's field is about 50 microtesla.

Turkish: Test zamanı uyarlaması

Briefly updating a trained model at inference time using only the incoming data itself. In reconstruction, aiming for consistency with the measured k-space can reduce the impact of domain shift.

Turkish: Uçuş zamanı anjiyografisi

A contrast-free vascular imaging method that relies on fresh, unsaturated blood flowing into the slice and appearing bright.

Turkish: Demet tabanlı uzaysal istatistik

A method that projects each subject's FA map onto a common white-matter skeleton for group comparison. It reduces sensitivity to registration errors and smoothing choices.

Turkish: Traktografi

Reconstructing white-matter pathways in 3D by following fibre directions estimated voxel by voxel. It can be deterministic or probabilistic. Results are indirect estimates of real axonal connections and can include false-positive pathways.

Turkish: Eğitim, doğrulama ve test kümeleri

Splitting data into three parts: to train the model, to choose its settings and to measure its final performance. If the same person's data appear in more than one set, performance looks better than it is.

Turkish: Dönüştürücü

A network architecture that models relations between parts of the input with an attention mechanism. It can capture context between distant regions. For images it works on small patches (vision transformer, ViT).

Turkish: Hızlı spin eko

A spin echo variant that collects several echoes after one excitation using a train of refocusing pulses. It speeds up the scan by the echo train length and is the standard way to acquire T2-weighted images.

A convolutional network that first downsamples the image to gather context, then upsamples to recover detail, with skip connections between the two paths. It was designed for medical image segmentation and also serves as the backbone of many reconstruction and diffusion models.

Turkish: Ultra yüksek alan

Field strengths of 7 tesla and above. They give higher SNR and resolution, while B₁ non-uniformity, SAR and susceptibility artifacts become more prominent.

Turkish: Belirsizlik kestirimi

Measuring how far a model's output can be trusted, at voxel or image level. It shows readers which regions deserve a careful look.

Turkish: Eksik örnekleme

Measuring only part of k-space to speed up a scan. The speed-up is given as the factor R. The sampling mask that decides which points are measured directly affects reconstruction quality and can itself be learned.

Turkish: Görme-dil modeli

A model that processes images and text together. It can be used to match radiology reports with images, answer questions about an image or draft a report.

Turkish: Voksel

The smallest volume element of an image. Its size is the field of view divided by the matrix, times slice thickness. Smaller voxels give less signal.

Turkish: Fermuar artefaktı

A zipper-like line running across the image. It usually comes from outside RF interference, for example a door left open or an electronic device brought into the room.