Deep Learning Classification of Lake Zooplankton

Plankton are effective indicators of environmental change and ecosystem health in freshwater habitats, but collection of plankton data using manual microscopic methods is extremely labor-intensive and expensive. Automated plankton imaging offers a promising way forward to monitor plankton communities with high frequency and accuracy in real-time. Yet, manual annotation of millions of images proposes a serious challenge to taxonomists. Deep learning classifiers have been successfully applied in various fields and provided encouraging results when used to categorize marine plankton images. Here, we present a set of deep learning models developed for the identification of lake plankton, and study several strategies to obtain optimal performances, which lead to operational prescriptions for users. To this aim, we annotated into 35 classes over 17900 images of zooplankton and large phytoplankton colonies, detected in Lake Greifensee (Switzerland) with the Dual Scripps Plankton Camera. Our best models were based on transfer learning and ensembling, which classified plankton images with 98% accuracy and 93% F1 score. When tested on freely available plankton datasets produced by other automated imaging tools (ZooScan, Imaging FlowCytobot, and ISIIS), our models performed better than previously used models. Our annotated data, code and classification models are freely available online.

FISH By Imaging Flow Cytometry in CLL for Diagnosis and MRD Assessment

Chronic lymphocytic leukaemia is a genetically heterogeneous disease with treatment and prognosis varying based on chromosomal abnormalities. These are detectable in up to 80% of cases when tested on the nuclei of interphase cells by fluorescence in situ hybridisation (FISH). Despite the clinical importance of FISH in management, as only up to 200 nuclei are generally assessed, it is not suitable for minimal residual disease (MRD) assessment. Since clinical decisions are based on detection thresholds of 10 -4, MRD assays are restricted to flow cytometry and molecular based assessment. Here we have explored the utility of a cutting-edge automated imaging flow cytometry method that incorporates cell immunophenotype and FISH ("immuno-flowFISH") to detect chromosomal abnormalities in CLL. Aims: Our aim was to determine the capability of immuno-flowFISH using imaging flow cytometry to detect del(17p) and +12 in CLL, and, the lowest limit of detection. We hypothesized that this integrated automated immuno-flowFISH method would be suitable for MRD assessment of CLL. Methods: Blood from 75 patients with CLL, at diagnosis or on therapy, was analysed. For MRD studies, cells from the CI cell line were spiked into normal blood at concentrations of 0.001 - 10%. After red cell lysis, samples were incubated with CD3, CD5 and CD19 fluorophore-conjugated antibodies (fluorophores: BV480, BV605, AF647). Following fixation and membrane permeabilization, DNA was denatured at 78 oC for 5 mins. FISH probes to 17p12 and centromeres of chromosomes 12 and 17 were added and hybridized for 24 hours at 42 oC. Nuclei were then stained with SYTOX AADvanced and up to 600,000 cells acquired on the Amnis ® ImageStream ®XMk II imaging flow cytometer. Digital images (x60 objective) and quantitative data derived from computational algorithms (IDEAS software) were used to assess FISH signals overlying cell nuclei. IDEAS was then used to assess the number and percent CD19/CD5-positive CLL cells with FISH abnormalities. Results: Between 10,000 and 600,000 cells (mean 60,000) were acquired. CLL (CD19/CD5-positive) and T- (CD3/CD5-positive) cells could be clearly identified by their immunophenotype and assessed individually for probe signals. FISH signals were identifiable on the digital images as specific "spots" overlying the SYTOX AADvanced nuclear stain. The IDEAS software could enumerate the number of FISH spots per cell and this was confirmed by quantitative mean channel fluorescence intensity for each probe. A chromosome 12 or 17 abnormality was detected in 23 of the 75 CLL cases. Of these, 10 cases had only one 17p signal (but 2 for the centromere of chromosome 17), indicative of del(17p). Del(17p) was detected in 2-35% of CD19/CD5-positive cells (i.e. 0.4-23% or 270-35,441 of all cells), the lowest seen in a patient on cytoreductive therapy. In 13/75 cases, there were 3 FISH signals for CEP12, consistent with trisomy 12 (+12) in 0.1-46% of all cells analysed; the lowest number of 0.1% was when 26 out of 26,000 cells analysed were CD19/CD5-positive and had +12. We also performed multi-FISH, incorporating CEP12, CEP17 and 17p probes together with the CD3, CD5 and CD19 antibodies. This required 7-fluorophores (antibodies, probes and nucleus) and confirmed the ability to detect del(17p) and chromosome 12 copy number simultaneously in a single analysis. Spiking of CI CLL cells into normal blood demonstrated that +12 could be detected to a lowest limit of 10 -5. In all analyses, CLL cells had normal diploid spots for the control CEP17 probe, and the CD3/CD5-positive T cells had dual signals for CEP12, CEP17 and 17p12 probes on numerical analysis and on digital imagery. Conclusion: This study of confirms that high-throughput automated imaging flow cytometry, integrating FISH and immunophenotyping, can detect chromosomal defects in CLL. The lowest limit of detection for a FISH-detectable abnormality was 10 -5. This high sensitivity and specificity exceeds current clinical practice (10 -4), and was achieved through the analysis of many thousands of cells and positive identification of CLL cells based on their phenotype. This immuno-flowFISH method does not require any prior cell separation and is automated. It adds a new dimension to chromosomal analysis in CLL, both at diagnosis and for MRD monitoring. The high precision and specificity of immuno-flowFISH illustrates that this has a real place as a new MRD assessment tool for CLL. Disclosures No relevant conflicts of interest to declare.

Deep Learning Classification of Lake Zooplankton

Plankton are effective indicators of environmental change and ecosystem health in freshwater habitats, but collection of plankton data using manual microscopic methods is extremely labor- intensive and expensive. Automated plankton imaging offers a promising way forward to monitor plankton communities with high frequency and accuracy in real-time. Yet, manual annotation of millions of images proposes a serious challenge to taxonomists. Deep learning classifiers have been successfully applied in various fields and provided encouraging results when used to categorize marine plankton images. Here, we present a set of deep learning models developed for the identification of lake plankton, and study several strategies to obtain optimal performances, which lead to operational prescriptions for users. To this aim, we annotated into 35 classes over 17900 images of zooplankton and large phytoplankton colonies, detected in Lake Greifensee (Switzerland) with the Dual Scripps Plankton Camera. Our best models were based on transfer learning and ensembling, which classified plankton images with 98% accuracy and 93% F1 score. When tested on freely available plankton datasets produced by other automated imaging tools (ZooScan, FlowCytobot and ISIIS), our models performed better than previously used models. Our annotated data, code and classification models are freely available online.

Comparison of an Automated Plate Assessment System (APAS Independence) and Artificial Intelligence (AI) to Manual Plate Reading of Methicillin-resistant and Methicillin-susceptible Staphylococcus aureus Chromagar Surveillance Cultures

Background: The Automated Plate Assessment System (APAS Independence) [Clever Culture System, Bäch, Switzerland] is an automated imaging station linked with interpretive software that detects target colonies of interest on chromogenic media and sorts samples as negative or presumptive positive. We evaluated the accuracy of the APAS to triage methicillin-resistant Staphylococcus aureus (MRSA) and S. aureus ) cultures using chromogenic media compared to human interpretation. Methods: Patient samples collected from the nares on Eswabs were plated to BD BBL™ CHROMagar™ MRSA II and BD BBL CHROMagar Staph aureus and allowed to incubate for 20-24 h at 37°C in non-CO2. Mauve colonies are suggestive of S. aureus and were confirmed with latex agglutination. Following incubation, samples were first interrogated by APAS before being read by a trained technologist blinded to the APAS interpretation. The triaging by both APAS and the technologists were compared for accuracy. Any discordant results required further analysis by a third reader. Results: Over a five-month period, 5,913 CHROMagar MRSA cultures were evaluated. Of those, 236 were read as concordantly positive, 5,525 were read as concordantly negative, and 152 required discordant analysis. Positive and negative percent agreements (PPA, NPA) were 100% and 97.3%, respectively. The APAS detected 5 positive cultures that were missed by manual reading, and determined to be true positives. In a separate analysis, 744 CHROMagar Staph aureus plates were read in parallel. Of these, 133 were concordantly positive, 585 were concordantly negative, and 26 required discordant analysis. PPA and NPA were 95.7% and 96.7%, respectively. Conclusion: This study confirmed the high sensitivity of digital image analysis by the APAS Independence such that negative cultures can be reliably reported without technologist intervention (NPV 100% for CHROMagar MRSA and 99.0% for CHROMagar Staph aureus). Triaging using the APAS Independence may provide great efficiencies in a laboratory with high throughput of MRSA and S. aureus surveillance cultures.

Images in Space and Time

Medical imaging diagnosis is mostly subjective, as it depends on medical experts. Hence, the service provided is limited by expert opinion variations and image complexity as well. However, with the increasing advancements in deep learning field, techniques are developed to help in the diagnosis and risk assessment processes. In this article, we survey different types of images in healthcare. A review of the concept and research methodology of Radiomics will highlight the potentials of integrated diagnostics. Convolutional neural networks can play an important role in next generations of automated imaging biomarker extraction and big data analytics systems. Examples are provided of what is already feasible today and also describe additional technological components required for successful clinical implementation.

A Systematic Approach in the Development of the Morphologically-Directed Raman Spectroscopy Methodology for Characterizing Nasal Suspension Drug Products

Demonstrating bioequivalence (BE) of nasal suspension sprays is a challenging task. Analytical tools are required to determine the particle size of the active pharmaceutical ingredient (API) and the structure of a relatively complex formulation. This study investigated the utility of the morphologically-directed Raman spectroscopy (MDRS) method to investigate the particle size distribution (PSD) of nasal suspensions. Dissolution was also investigated as an orthogonal technique. Nasal suspension formulations containing different PSD of mometasone furoate monohydrate (MFM) were manufactured. The PSD of the MFM batches was characterized before formulation manufacture using laser diffraction and automated imaging. Upon formulation manufacture, the droplet size, single actuation content, spray pattern, plume geometry, the API dissolution rate, and the API PSD by MDRS were determined. A systematic approach was utilized to develop a robust method for the analysis of the PSD of MFM in Nasonex® and four test formulations containing the MFM API with different particle size specifications. Although the PSD between distinct techniques cannot be directly compared due to inherent differences between these methodologies, the same trend is observed for three out of the four batches. Dissolution analysis confirmed the trend observed by MDRS in terms of PSD. For suspension-based nasal products, MDRS allows the measurement of API PSD which is critical for BE assessment. This approach has been approved for use in lieu of a comparative clinical endpoint BE study [1]. The correlation observed between PSD and dissolution rate extends the use of dissolution as a critical analytical tool demonstrating BE between test and reference products.

Clustering and Erratic Movement Patterns of Syringe-Injected versus Mosquito-Inoculated Malaria Sporozoites Underlie Decreased Infectivity

ABSTRACT Malaria vaccine candidates based on live, attenuated sporozoites have led to high levels of protection. However, their efficacy critically depends on the sporozoites’ ability to reach and infect the host liver. Administration via mosquito inoculation is by far the most potent method for inducing immunity but highly impractical. Here, we observed that intradermal syringe-injected Plasmodium berghei sporozoites (syrSPZ) were 3-fold less efficient in migrating to and infecting mouse liver than mosquito-inoculated sporozoites (msqSPZ). This was related to a clustered dermal distribution (2-fold-decreased median distance between syrSPZ and msqSPZ) and, more importantly, a 1.4-fold (significantly)-slower and more erratic movement pattern. These erratic movement patterns were likely caused by alteration of dermal tissue morphology (>15-μm intercellular gaps) due to injection of fluid and may critically decrease sporozoite infectivity. These results suggest that novel microvolume-based administration technologies hold promise for replicating the success of mosquito-inoculated live, attenuated sporozoite vaccines. IMPORTANCE Malaria still causes a major burden on global health and the economy. The efficacy of live, attenuated malaria sporozoites as vaccine candidates critically depends on their ability to migrate to and infect the host liver. This work sheds light on the effect of different administration routes on sporozoite migration. We show that the delivery of sporozoites via mosquito inoculation is more efficient than syringe injection; however, this route of administration is highly impractical for vaccine purposes. Using confocal microscopy and automated imaging software, we demonstrate that syringe-injected sporozoites do cluster, move more slowly, and display more erratic movement due to alterations in tissue morphology. These findings indicate that microneedle-based engineering solutions hold promise for replicating the success of mosquito-inoculated live, attenuated sporozoite vaccines.

Democratising “Microscopi”: a 3D printed automated XYZT fluorescence imaging system for teaching, outreach and fieldwork

Commercial fluorescence microscope stands and fully automated XYZt fluorescence imaging systems are generally beyond the limited budgets available for teaching and outreach. We have addressed this problem by developing “Microscopi”, an accessible, affordable, DIY automated imaging system that is built from 3D printed and commodity off-the-shelf hardware, including electro-mechanical, computer and optical components. Our design features automated sample navigation and image capture with a simple web-based graphical user interface, accessible with a tablet or other mobile device. The light path can easily be switched between different imaging modalities. The open source Python-based control software allows the hardware to be driven as an integrated imaging system. Furthermore, the microscope is fully customisable, which also enhances its value as a learning tool. Here, we describe the basic design and demonstrate imaging performance for a range of easily sourced specimens.