Robust Feature-Sample Linear Discriminant Analysis for Brain Disorders Diagnosis
Authors: Ehsan Adeli-Mosabbeb, Kim-Han Thung, Le An, Feng Shi, Dinggang Shen
NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test our algorithm on one synthetic and two brain neurodegenerative databases (particularly for Parkinson s disease and Alzheimer s disease). The results demonstrate that our method outperforms all baseline and state-of-the-art methods, in terms of both accuracy and the area under the ROC curve. and 3 Experiments We compare our method with several baseline and state-of-the-art methods in three different scenarios. The first experiment is on synthetic data... The next two experiments are conducted for neurodegenerative brain disorders diagnosis. |
| Researcher Affiliation | Academia | Department of Radiology and BRIC University of North Carolina at Chapel Hill, NC, 27599, USA {eadeli,khthung,le_an,fengshi,dgshen}@med.unc.edu |
| Pseudocode | Yes | Algorithm 1 RFS-LDA optimization algorithm. Input: X = [Xtr Xtst], Ytr, parameters η, λ1, λ2, ρ and γ. Initialization: D0 = [Xtr Xtst], ˆD0 = [Xtr; 1 ],β0 = Ytr( ˆD0) ( ˆD0( ˆD0) + γI), E0 = 0, L 0 1 = X/ X 2, L 0 2 = Xtr/ Xtr 2, L 0 3 = β0/ β0 2, µ1 = d N 4 X 1, µ2 = d Ntr 4 Xtr 1, µ3 = dc 4 β0 1. 1: k 0 2: repeat Main optimization loop 3: t 0, ˆβ 0 = βk Update β 4: repeat 5: i, j {0, . . . , Ntr 1}, i = j, ˆαij 0 and ˆαii 1/ q (yk i ˆβt ˆdk i )2 + 0.0001 6: ˆβ t+1 Ytrˆαˆα ( ˆDk) + µ3(Bk L k 3 ) ˆDkˆαˆα ( ˆDk) + γI , t t + 1 7: until ˆβt 1 ˆβt F/( ˆβt 1 F ˆβt F) < 0.001 or t > 100 8: βk+1 ˆβ t. 9: ˆDk+1 ηˆα (βk+1) βk+1ˆα + µk 2I 1 ηˆα (βk+1) Ytr L k 2 + µk 2[Dk tr; 1 ] Update ˆD 10: Dk+1 D1/(µk 1 + µk 2 ) L k 1 + µk 1(X Ek) + [L k 2 + µk 2 ˆDk+1](1:Ntr,:) 0 Update D 11: Ek+1 Sλ1/µk 1 (X Dk+1 + L k 1 /µk 1) Update E 12: Bk+1 Sλ2/µk 3 (βk+1 + L k 3 ) Update B 13: L k+1 1 L k 1 + µk 1(X Dk+1 Ek+1) Update multipliers and parameters 14: L k+1 2 L k 2 + µk 2( ˆD [Dk+1 tr ; 1 ]), L k+1 3 L k 3 + µk 3(β B) 15: µk+1 1 min(ρµk 1, 109), µk+1 2 min(ρµk 2, 109), µk+1 3 min(ρµk 3, 109) 16: k k + 1 17: until X Dk Ek F/ X F < 10 8 and ˆ Dk [Dk tr; 1 ] F/ ˆ Dk F < 10 8 and βk Bk F/ βk F < 10 8 Output: β, D, E and Ytst = βXtst. |
| Open Source Code | No | The paper provides URLs for the ADNI and PPMI databases used as datasets but does not provide any link or explicit statement about releasing the source code for the proposed RFS-LDA method. |
| Open Datasets | Yes | Parts of the data used in preparation of this article were obtained from the Alzheimer s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.ucla.edu). and The first set of data used in this paper is obtained from the Parkinson s progression markers initiative (PPMI) database2 [23]. |
| Dataset Splits | Yes | For the choice of parameters, the best parameters are selected through an inner 10-fold cross validation on the training data, for all the competing methods. and Note that all the reported results are obtained through 10-fold cross-validation. and Table 1 shows the diagnosis accuracy of the proposed technique (RFS-LDA) in comparisons with different baseline and state-of-the-art methods, using a 10-fold cross-validation strategy. |
| Hardware Specification | No | The paper mentions '3T SIEMENS MAGNETOM Trio Tim syngo scanners' for MRI data acquisition, but does not provide specific details (like CPU, GPU models, or memory) of the computing hardware used to run the experiments or train the models. |
| Software Dependencies | No | The paper mentions software packages like 'HAMMER' [25, 30] and 'FSL package' [33] used for data preprocessing. However, it does not specify the version numbers for these or any other software dependencies required to replicate the experiments. |
| Experiment Setup | Yes | For the proposed method, the parameters are set with a same strategy as in [15]: λ1 = Λ1/( p min(d, N)), λ2 = Λ2/ d, ηk = Λ3 X / Ytr βk ˆDk 2 F, and ρ (controlling the {µ}s in the algorithm) is set to 1.01. We have set Λ1, Λ2, Λ3 and γ through inner cross validation, and found that all set to 1 yields to reasonable results across all datasets. |