Visual interpretability of image-based classification models by generative latent space disentanglement applied to in vitro fertilization.

The success of deep learning in identifying complex patterns exceeding human intuition comes at the cost of interpretability. Non-linear entanglement of image features makes deep learning a "black box" lacking human meaningful explanations for the models' decision. We present DISCOVER, a generative model designed to discover the underlying visual properties driving image-based classification models. DISCOVER learns disentangled latent representations, where each latent feature encodes a unique classification-driving visual property. This design enables "human-in-the-loop" interpretation by generating disentangled exaggerated counterfactual explanations. We apply DISCOVER to interpret classification of in vitro fertilization embryo morphology quality. We quantitatively and systematically confirm the interpretation of known embryo properties, discover properties without previous explicit measurements, and quantitatively determine and empirically verify the classification decision of specific embryo instances. We show that DISCOVER provides human-interpretable understanding of "black box" classification models, proposes hypotheses to decipher underlying biomedical mechanisms, and provides transparency for the classification of individual predictions.

Correlative analyses of nitric oxide generation rates and nitric oxide synthase levels in individual cells using a modular cell-retaining device.

Nitric oxide (NO) is recognized as one of the major immune system agents involved in the pathogenesis and control of various diseases that may benefit from novel drug development, by exploiting NO signaling pathways and targets. This calls for detection of both intracellular levels of NO and expression of its synthesizing enzymes (NOS) in individual, intact, living cells. Such measurements are challenging, however, due to short half-life, low and fluctuating concentrations of NO, cellular heterogeneity, and inability to trace the same cells over time. The current study presents a device and methodology for correlative analysis of NO generation rates and NOS levels in the same individual cells, utilizing fluorescent imaging followed by immunohistochemistry (IHC). U937 promonocyte cell populations demonstrated significant heterogeneity in their baseline levels, in NO-generation kinetics, and in their response rates to stimuli. Individual cell analysis exposed cell subgroups which showed enhanced NO production upon stimulation, concomitantly with significant up-regulation of inducible NOS (iNOS) levels. Exogenous NO modulated the expression of iNOS in nondifferentiated cells within 1 h, in a dose-dependent manner, while treatment with lysophosphatidylcholine (LPC) enhanced the expression of iNOS, demonstrating a nondependence on NO production.

A polymer microstructure array for the formation, culturing, and high throughput drug screening of breast cancer spheroids.

Multicellular spheroid models have been recognized as superior to monolayer cell cultures in antitumor drug screening, but their commercial adaptation in the pharmaceutical industry has been delayed, primarily due to technological limitations. The current study presents a new spheroid culture platform that addresses these technical restrictions. The new culturing device is based on a multiwell plate equipped with a glass bottom patterned with an array of UV adhesive microchambers. Each microchamber is designed to accommodate a single spheroid. The system facilitates the simultaneous creation and culturing of a large number of spheroids, as well as screening their response to antitumor drugs. The volume of the spheroids is easily controlled by seeding density. The location of each spheroid is preserved in the same microchamber throughout its growth, treatment with soluble agents, and imaging. The growth ratio parameter, a non-intrusive size analysis of the same spheroid before and after exposure to drugs, was found to be a sensitive indicator for the reaction of MCF7 breast cancer spheroids to cytotoxic drugs. This feature helps reveal the heterogeneity within the spheroid population during the formation process and their drug response, and provides an opportunity to detect specific, highly active or drug-resistant spheroid sub-groups. The advantages of this spheroid-based system make it an efficient drug-screening tool that may be valuable to related fields of research and clinical applications.

Polymer live-cell array for real-time kinetic imaging of immune cells.

Direct quantitative experimental investigations of the function of lymphocytes and other immune cells are challenging due to the cell mobility and the complexity of intercellular communications. In order to facilitate such investigations, an in vitro system is required that is noninvasive and provides kinetic data on cellular responses to challenges such as drug treatments. The present work reports the development of a disposable, inexpensive polymer-made device, the Polymer Live Cell Array (PLCA), for real-time, kinetic analysis of immune cells. The PLCA proved to be optically and biologically compatible, thus individual immune cells can be observed and treated independently without being tethered. The cells share a common space which facilitates cellular communications via secreted molecules or via direct intercellular interactions. These properties facilitate real-time, non-intrusive, repeated measurements of immune cells under multiple experimental treatments.