Research management with papers
Tencent's pretrain model for medical image analysis, trained and tested on segmentation challanges, but pretrained on Medical data only. seems promising, need to check.
Learning from minimal labeled + large unlabeled data by using aumentation and sharpening - seems intersting to segmentation variant - Need to try!
Simultaneous Truth and Performance Level Estimation (STAPLE): An Algorithm for the Validation of Image Segmentation paper code1 code-orig
Using EM variant alogorithm for estimation of GT with varirty of segmentations (/algorithms). The algorithm generates simultanously (M-step) Probablistic map of segmentation and then evaluate(E-step) the given segmentations vs the map. with extension to multiclass segmentation(partD) and global features(partE). Note that the algorithm assumes pixels are iid so no matter 2D or 3D. evalutaion was done on brain MRI and prostate MRI. extension algorithm links
Adaloss: Adaptive Loss Function for Landmark Localization paper
Train landmark detector that increase the accuracy ( by reducing SDof gaussion for landmark detection) in train time adaptive to each class. Currently State-of-the-art in Face landmark detection (AFLW,300W)
previous StoA, addressing style varaiations of face landmark by CycleGAN generation, may be useful for robustness.
Symmetry line(s) detection - Doesnt seems to work on 2D or I did not understand the output (R/theta?)
Symmetry line detection. Seems to work with Seg * Image on our data.
Cerebral biometry in fetal magnetic resonance imaging: new reference data (Garel) paper
Reference data for MRI biometry.
Prenatal Brain MR Imaging: Reference Linear Biometric Centiles between 20 and 24 Gestational Weeks paper
Also have Bland-Altman plots and tables with numerical values for intra-doctor aggreement on linear measurement for fetuses (20-24).
Normative biometry of the fetal brain using magnetic resonance imaging(Kyrkyapodoulu) paper
Explanataion of how we need to measure bometrics and reference excel with cacluations
Accuracy of transverse cerebellar diameter measurement by ultrasonography in the evaluation of foetal age paper
Shows that TCD is better predictor than other linear mesurement for IUGR. Note that this done on US (which have more noise than MRI) but may (and we can see it by the variance) be true also for MRI. The foetal typo is original.
Assessment of MRI-Based Automated Fetal Cerebral Cortical Folding Measures in Prediction of Gestational Age in the Third Trimesterpaper
Estimation of brain age using volume and cortical folding (sulcuses). evaluate 8 types of measuring sulcuses and take the best fit. take care that they didnt have a lot of data but still they got 98% R-square.
Deep Learning with Attention to Predict Gestational Age of the Fetal Brain paper
Using Unet+Attention on brain to estimate GA. Private DB of ~700 fetuses and straightforward(Medical) approach. By using multi-view(Cor+Ax+Sag) they got better result.
A NEW SYMMETRY-BASED METHOD FOR MID-SAGITTAL PLANE EXTRACTION IN NEUROIMAGES paper code - Take cre , only exe
Find mis dagittal plane (MSP) in T1 brain images automatically, by using symetry-to-plane. caluclating symmetry by correletaion between images in two side of plane and minimizing.
An Efficient Automatic Midsagittal Plane Extraction in Brain MRI paper
Faster(1s for 3D) and geometric, still need to understand how it works
An Overview on Assessing Agreement with Continuous Measurements paper
Overview
A graphical method for assessing agreement with the mean between multiple observers using continuous measures paper
Estimator for inter-observer variability for more than 2 observers, take care that is is not exactly as Balnd-Altman results.
In test, The code does not work on my GPU (CUDA Error in patchBasedPSFReconstructionKernel(), line 132: too many resources requested for launch)
For me, There is need for dataset (They use iFind) So I will try to reconstruct one from equivoxel T2 (Preterm) and resample.
Robust Super-resolution Volume Reconstruction from Slice Acquisitions: Application to Fetal Brain MRI paper
Simon's paper - still trying to find it's code.
Slice-to-volume medical image registration: a survey paper
Need to move over it again
An automated framework for localization, segmentation and super-resolution reconstruction of fetal brain MRI paper code
Full pipeline of localization and reconstruction for fetal, trained on SSFSE. seems promising.
Did try it, it doesnt recontruct well. "wave" artifacts, It may be effect of optimization paramteres
Learning-Based Compressive MRI paper
Learning of sampling pattern and reconstruction algorithm for MRI simultaneously. 4-fold, 1D/2D sampling.
Using cascaded CNN and GAN for perceptual refinement for heart MRI. 3 to 9 fold, 1D variable density sampling. Shows that SNR doesn’t necessarily correlate with image quality and radiological interpretability.
- We try to run this code and the results wasnt good - need to understand why.
Deep Learning Based Multi-Modal Fusion for Fast MR Reconstruction paper
close idea, but for T1+ Undersampled T2 -> T2. take care that T2 has higher SNR than FLAIR inherently.
-
Brats2018 - Brain Lesion Segmenation - HGG and LGG Link [v]
Constrast : T1, T1Gd, T2, FLAIR
Mode : Preprocessed
Type : MHA(?)
Count : 210 HGG + 75 LGG = 285 Patients, No control
Annotations : Lesions - split to components -
ABIDE [v] Constrast : T1-MPRAGE
Count : 1109
Type : NiFTi
[Place]
- FastMRI - Knee [x]
| GE | Siemens | Philips |
|---|---|---|
| ssFSE | HASTE | |
| FIESTA | TruFI |