Journal: Scientific Reports

Article Title: An intelligent learning system based on electronic health records for unbiased stroke prediction

doi: 10.1038/s41598-024-73570-x

Figure Lengend Snippet: Performance comparison with state-of-the-art (SOTA).

Article Snippet: Sailasya et al. , No , No , Low , Kaggle , Naive Bayes , 82.00%.

Techniques: Comparison, Plasmid Preparation

Journal: Scientific Reports

Article Title: Deep Analysis of Mitochondria and Cell Health Using Machine Learning

doi: 10.1038/s41598-018-34455-y

Figure Lengend Snippet: Overview of MitoMo software pipeline. Video frames of labeled mitochondria were captured and loaded into the MitoMo software. The videos were segmented and declumped, and 21 morphological, intensity, texture, and motion features were extracted for each frame. The selected features were then fed into a previously trained library of the three mitochondrial morphologies (punctate, networked, and swollen). The segmented mitochondria from the test videos (frames) were automatically morphologically classified using K-nearest neighbor (KNN) and Naïve Bayes. The resulting data were plotted into graphs, depicting the percentage of punctate, networked, and swollen mitochondria. The software can also perform motion analysis by computing the magnitude and orientation of the gradient vectors, and the net or directional motion can be plotted for the population of mitochondria in each cell. Motion analysis can be performed for the entire mitochondrial population within a cell, individual mitochondria, and/or morphological classes of mitochondria. Texture features, which are indicative of mitochondrial complexity and organization, can be extracted, Validation was performed to ensure accuracy of the segmentation, morphological classification, and motion analysis. Extracted features can be combined to perform health classification during low-level stress that is otherwise not detectable.

Article Snippet: The extracted features were fed to supervised learning algorithms KNN and Naïve Bayes written on the MATLAB platform.

Techniques: Software, Labeling, Biomarker Discovery

Journal: Scientific Reports

Article Title: Deep Analysis of Mitochondria and Cell Health Using Machine Learning

doi: 10.1038/s41598-018-34455-y

Figure Lengend Snippet: MitoMo validation and comparison of our motion analysis to optical flow. Segmentation validation: ( A ) Manually drawn segmentation (green) using ImageJ was compared to ( B ) segmentation performed by MitoMo. ( C ) The segmentation accuracy (area of overlaps) of the manually labeled versus automatically segmented mitochondria (MitoMo and CellProfiler) were not significantly different using one-way ANOVA with Dunnett’s post hoc test. ( D – E ) The morphological classification was validated by comparing manually labeled mitochondria ( D ) against automatic classification ( E ). ( F ) There was up to 88% accuracy with Naïve Bayes and 80% accuracy with KNN in the automatic classification of mitochondrial morphology in 1761 trials using the training data. There was up to 89% accuracy with Naïve Bayes and 91% with KNN in the automatic classification of the test set. ( G – L ) Images showing motion analysis validation for both magnitude (blue, green cyan images) and direction (images with red arrows). Each magnitude image is a synthetic image of unique shape and change in intensity overlaid with a motion vector. Blue is object from frame 1, green is object from frame 2, and cyan is overlap of the two frames. Column 1 in ( G – L ) shows summed MitoMo vector (indicated by arrow) overlaid onto color coded image. Column 2 is the normalized difference image with MitoMo motion vectors. Column 3 shows the summed Lucas-Kanade vector (indicated by arrow) overlaid onto the color-coded images. Column 4 is the normalized difference image with Lucas-Kanade motion vectors. ( G ) Solid dot with no change in intensity. ( H ) Dot with uneven intensity. ( I ) Disc with uneven intensity that fades to half the brightness by the second frame. ( J ) Solid rectangle with no change in intensity. ( K ) Rectangle with uneven intensity. ( L ) Rectangle with uneven intensity that fades to half the brightness by the second frame. ( M ) Table summarizes the accuracy of angle and magnitude for MitoMo versus optical flow. The angle and magnitude accuracies were statistically compared using t-test and Chi-squared tests, and the more accurate software was highlighted in yellow.

Article Snippet: The extracted features were fed to supervised learning algorithms KNN and Naïve Bayes written on the MATLAB platform.

Techniques: Biomarker Discovery, Comparison, Labeling, Plasmid Preparation, Software