A multi-path 2.5 dimensional convolutional neural network system for segmenting stroke lesions in brain MRI images.

Automatic identification of brain lesions from magnetic resonance imaging (MRI) scans of stroke survivors would be a useful aid in patient diagnosis and treatment planning. It would also greatly facilitate the study of brain-behavior relationships by eliminating the laborious step of having a human expert manually segment the lesion on each brain scan. We propose a multi-modal multi-path convolutional neural network system for automating stroke lesion segmentation. Our system has nine end-to-end UNets that take as input 2-dimensional (2D) slices and examines all three planes with three different normalizations. Outputs from these nine total paths are concatenated into a 3D volume that is then passed to a 3D convolutional neural network to output a final lesion mask. We trained and tested our method on datasets from three sources: Medical College of Wisconsin (MCW), Kessler Foundation (KF), and the publicly available Anatomical Tracings of Lesions After Stroke (ATLAS) dataset. To promote wide applicability, lesions were included from both subacute (1 to 5 weeks) and chronic ( >  3 months) phases post stroke, and were of both hemorrhagic and ischemic etiology. Cross-study validation results (with independent training and validation datasets) were obtained to compare with previous methods based on naive Bayes, random forests, and three recently published convolutional neural networks. Model performance was quantified in terms of the Dice coefficient, a measure of spatial overlap between the model-identified lesion and the human expert-identified lesion, where 0 is no overlap and 1 is complete overlap. Training on the KF and MCW images and testing on the ATLAS images yielded a mean Dice coefficient of 0.54. This was reliably better than the next best previous model, UNet, at 0.47. Reversing the train and test datasets yields a mean Dice of 0.47 on KF and MCW images, whereas the next best UNet reaches 0.45. With all three datasets combined, the current system compared to previous methods also attained a reliably higher cross-validation accuracy. It also achieved high Dice values for many smaller lesions that existing methods have difficulty identifying. Overall, our system is a clear improvement over previous methods for automating stroke lesion segmentation, bringing us an important step closer to the inter-rater accuracy level of human experts.

Reconstruction of large phylogenetic trees: a parallel approach.

Reconstruction of phylogenetic trees for very large datasets is a known example of a computationally hard problem. In this paper, we present a parallel computing model for the widely used Multiple Instruction Multiple Data (MIMD) architecture. Following the idea of divide-and-conquer, our model adapts the recursive-DCM3 decomposition method [Roshan, U., Moret, B.M.E., Williams, T.L., Warnow, T, 2004a. Performance of suptertree methods on various dataset decompositions. In: Binida-Emonds, O.R.P. (Eds.), Phylogenetic Supertrees: Combining Information to Reveal the Tree of Life, vol. 3 of Computational Biology, Kluwer Academics, pp. 301-328; Roshan, U., Moret, B.M.E., Williams, T.L., Warnow, T., 2004b. Rec-I-DCM3: A Fast Algorithmic Technique for reconstructing large phylogenetic trees, Proceedings of the IEEE Computational Systems Bioinformatics Conference (ICSB)] to divide datasets into smaller subproblems. It distributes computation load over multiple processors so that each processor constructs subtrees on each subproblem within a batch in parallel. It finally collects the resulting trees and merges them into a supertree. The proposed model is flexible as far as methods for dividing and merging datasets are concerned. We show that our method greatly reduces the computational time of the sequential version of the program. As a case study, our parallel approach only takes 22.1h on four processors to outperform the best score to date (Found at 123.7h by the Rec-I-DCM3 program [Roshan, U., Moret, B.M.E., Williams, T.L., Warnow, T, 2004a. Performance of suptertree methods on various dataset decompositions. In: Binida-Emonds, O.R.P. (Eds.), Phylogenetic Supertrees: Combining Information to Reveal the Tree of Life, vol. 3 of Computational Biology, Kluwer Academics, pp. 301-328; Roshan, U., Moret, B.M.E., Williams, T.L., Warnow, T., 2004b. Rec-I-DCM3: A Fast Algorithmic Technique for reconstructing large phylogenetic trees, Proceedings of the IEEE Computational Systems Bioinformatics Conference (ICSB)] on one dataset. Developed with the standard message-passing library, MPI, the program can be recompiled and run on any MIMD systems.

Rec-I-DCM3: a fast algorithmic technique for reconstructing large phylogenetic trees.

Phylogenetic trees are commonly reconstructed based on hard optimization problems such as maximum parsimony (MP) and maximum likelihood (ML). Conventional MP heuristics for producing phylogenetic trees produce good solutions within reasonable time on small datasets (up to a few thousand sequences), while ML heuristics are limited to smaller datasets (up to a few hundred sequences). However, since MP (and presumably ML) is NP-hard, such approaches do not scale when applied to large datasets. In this paper, we present a new technique called Recursive-Iterative-DCM3 (Rec-I-DCM3), which belongs to our family of Disk-Covering Methods (DCMs). We tested this new technique on ten large biological datasets ranging from 1,322 to 13,921 sequences and obtained dramatic speedups as well as significant improvements in accuracy (better than 99.99%) in comparison to existing approaches. Thus, high-quality reconstructions can be obtained for datasets at least ten times larger than was previously possible.