Medical decision support system using weakly-labeled lung CT scans.

Purpose: Determination and development of an effective set of models leveraging Artificial Intelligence techniques to generate a system able to support clinical practitioners working with COVID-19 patients. It involves a pipeline including classification, lung and lesion segmentation, as well as lesion quantification of axial lung CT studies. Approach: A deep neural network architecture based on DenseNet is introduced for the classification of weakly-labeled, variable-sized (and possibly sparse) axial lung CT scans. The models are trained and tested on aggregated, publicly available data sets with over 10 categories. To further assess the models, a data set was collected from multiple medical institutions in Colombia, which includes healthy, COVID-19 and patients with other diseases. It is composed of 1,322 CT studies from a diverse set of CT machines and institutions that make over 550,000 slices. Each CT study was labeled based on a clinical test, and no per-slice annotation took place. This enabled a classification into Normal vs. Abnormal patients, and for those that were considered abnormal, an extra classification step into Abnormal (other diseases) vs. COVID-19. Additionally, the pipeline features a methodology to segment and quantify lesions of COVID-19 patients on the complete CT study, enabling easier localization and progress tracking. Moreover, multiple ablation studies were performed to appropriately assess the elements composing the classification pipeline. Results: The best performing lung CT study classification models achieved 0.83 accuracy, 0.79 sensitivity, 0.87 specificity, 0.82 F1 score and 0.85 precision for the Normal vs. Abnormal task. For the Abnormal vs COVID-19 task, the model obtained 0.86 accuracy, 0.81 sensitivity, 0.91 specificity, 0.84 F1 score and 0.88 precision. The ablation studies showed that using the complete CT study in the pipeline resulted in greater classification performance, restating that relevant COVID-19 patterns cannot be ignored towards the top and bottom of the lung volume. Discussion: The lung CT classification architecture introduced has shown that it can handle weakly-labeled, variable-sized and possibly sparse axial lung studies, reducing the need for expert annotations at a per-slice level. Conclusions: This work presents a working methodology that can guide the development of decision support systems for clinical reasoning in future interventionist or prospective studies.

Unsupervised fuzzy binning of metagenomic sequence fragments on three-dimensional Barnes-Hut t-Stochastic Neighbor Embeddings.

Shotgun metagenomic studies attempt to reconstruct population genome sequences from complex microbial communities. In some traditional genome demarcation approaches, high-dimensional sequence data are embedded into two-dimensional spaces and subsequently binned into candidate genomic populations. One such approach uses a combination of the Barnes-Hut approximation and the $t -$Stochastic Neighbor Embedding (BH-SNE) algorithm for dimensionality reduction of DNA sequence data pentamer profiles; and demarcation of groups based on Gaussian mixture models within humanimposed boundaries. We found that genome demarcation from three-dimensional BH-SNE embeddings consistently results in more accurate binnings than 2-D embeddings. We further addressed the lack of a priori population number information by developing an unsupervised binning approach based on the Subtractive and Fuzzy c-means (FCM) clustering algorithms combined with internal clustering validity indices. Lastly, we addressed the subject of shared membership of individual data objects in a mixed community by assigning a degree of membership to individual objects using the FCM algorithm, and discriminated between confidently binned and uncertain sequence data objects from the community for subsequent biological interpretation. The binning of metagenome sequence fragments according to thresholds in the degree of membership opens the door for the identification of horizontally transferred elements and other genomic regions of uncertain assignment in which biologically meaningful information resides. The reported approach improves the unsupervised genome demarcation of populations within complex communities, increases the confidence in the coherence of the binned elements, and enables the identification of evolutionary processes ignored in hard-binning approaches in shotgun metagenomic studies.

Bio-EdIP: An automatic approach for in vitro cell confluence images quantification.

BACKGROUND AND OBJECTIVES: Cell imaging is a widely-employed technique to analyze multiple biological processes. Therefore, simple, accurate and quantitative tools are needed to understand cellular events. For this purpose, Bio-EdIP was developed as a user-friendly tool to quantify confluence levels using cell culture images. METHODS: The proposed algorithm combines a pre-processing step with subsequent stages that involve local processing techniques and a morphological reconstruction-based segmentation algorithm. Segmentation performance was assessed in three constructed image sets, comparing F-measure scores and AUC values (ROC analysis) for Bio-EdIP, its previous version and TScratch. Furthermore, segmentation results were compared with published algorithms using eight public benchmarks. RESULTS: Bio-EdIP automatically segmented cell-free regions from images of in vitro cell culture. Based on mean F-measure scores and ROC analysis, Bio-EdIP conserved a high performance regardless of image characteristics of the constructed dataset, when compared with its previous version and TScratch. Although acquisition quality of the public dataset affected Bio-EdIP segmentation, performance was better in two out of eight public sets. CONCLUSIONS: Bio-EdIP is a user-friendly interface, which is useful for the automatic analysis of confluence levels and cell growth processes using in vitro cell culture images. Here, we also presented new manually annotated data for algorithms evaluation.