Search

Separating and labeling each nuclear instance (instance-aware segmentation) is the key challenge in nuclear image segmentation. Deep Convolutional Neural Networks have been demonstrated to solve nuclear image segmentation tasks across different imaging modalities, but a systematic comparison on complex immunofluorescence images has not been performed. Deep learning based segmentation requires annotated datasets for training, but annotated fluorescence nuclear image datasets are rare and of limited size and complexity. In this work, we evaluate and compare the segmentation effectiveness of multiple deep learning architectures (U-Net, U-Net ResNet, Cellpose, Mask R-CNN, KG instance segmentation) and two conventional algorithms (Iterative h-min based watershed, Attributed relational graphs) on complex fluorescence nuclear images of various types. We propose and evaluate a novel strategy to create artificial images to extend the training set. Results show that instance-aware segmentation architectures and Cellpose outperform the U-Net architectures and conventional methods on complex images in terms of F1 scores, while the U-Net architectures achieve overall higher mean Dice scores. Training with additional artificially generated images improves recall and F1 scores for complex images, thereby leading to top F1 scores for three out of five sample preparation types. Mask R-CNN trained on artificial images achieves the overall highest F1 score on complex images of similar conditions to the training set images while Cellpose achieves the overall highest F1 score on complex images of new imaging conditions. We provide quantitative results demonstrating that images annotated by under-graduates are sufficient for training instance-aware segmentation architectures to efficiently segment complex fluorescence nuclear images.

In order to develop machine learning and deep learning models that take into account the guidelines and principles of trustworthy AI, a novel information theoretic approach is introduced in this article. A unified approach to privacy-preserving interpretable and transferable learning is considered for studying and optimizing the trade-offs between the privacy, interpretability, and transferability aspects of trustworthy AI. A variational membership-mapping Bayesian model is used for the analytical approximation of the defined information theoretic measures for privacy leakage, interpretability, and transferability. The approach consists of approximating the information theoretic measures by maximizing a lower-bound using variational optimization. The approach is demonstrated through numerous experiments on benchmark datasets and a real-world biomedical application concerned with the detection of mental stress in individuals using heart rate variability analysis. [ABSTRACT FROM AUTHOR]

Medical Assisted Reproduction proved its efficacy to treat the vast majority forms of infertility. One of the key procedures in this treatment is the selection and transfer of the embryo with the highest developmental potential. To assess this potential, clinical embryologists routinely work with static images (morphological assessment) or short video sequences (time-lapse annotation). Recently, Artificial Intelligence models were utilized to support the embryo selection procedure. Even though they have proven their great potential in different in vitro fertilization settings, there is still considerable room for improvement. To support the advancement of algorithms in this research field, we built a dataset consisting of static blastocyst images and additional annotations. As such, Gardner criteria annotations, depicting a morphological blastocyst rating scheme, and collected clinical parameters are provided. The presented dataset is intended to be used to train deep learning models on static morphological images to predict Gardner's criteria and clinical outcomes such as live birth. A benchmark of human expert's performance in annotating Gardner criteria is provided. [ABSTRACT FROM AUTHOR]

In the last decade, industry's demand for deep learning (DL) has increased due to its high performance in complex scenarios. Due to the DL method's complexity, experts and non-experts rely on blackbox software packages such as Tensorflow and Pytorch. The frameworks are constantly improving, and new versions are released frequently. As a natural process in software development, the released versions contain improvements/changes in the methods and their implementation. Moreover, versions may be bug-polluted, leading to the model performance decreasing or stopping the model from working. The aforementioned changes in implementation can lead to variance in obtained results. This work investigates the effect of implementation changes in different major releases of these frameworks on the model performance. We perform our study using a variety of standard datasets. Our study shows that users should consider that changing the framework version can affect the model performance. Moreover, they should consider the possibility of a bug-polluted version before starting to debug source code that had an excellent performance before a version change. This also shows the importance of using virtual environments, such as Docker, when delivering a software product to clients. [ABSTRACT FROM AUTHOR]

Clinical decision-making in cancer and other diseases relies on timely and cost-effective genome-wide testing. Classical bioinformatic algorithms, such as Rawcopy, can support genomic analysis by calling genomic breakpoints and copy-number variations (CNVs), but often require manual data curation, which is error prone, time-consuming, and thus substantially increasing costs of genomic testing and hampering timely delivery of test results to the treating physician. We aimed to investigate whether deep learning algorithms can be used to learn from genome-wide single-nucleotide polymorphism array (SNPa) data and improve state-of-the-art algorithms. We developed, applied, and validated a novel deep neural network (DNN), DeepSNP. A manually curated data set of 50 SNPa analyses was used as truth-set. We show that DeepSNP can learn from SNPa data and classify the presence or absence of genomic breakpoints within large genomic windows with high precision and recall. DeepSNP was compared with well-known neural network models as well as with Rawcopy. Moreover, the use of a localization unit indicates the ability to pinpoint genomic breakpoints despite their exact location not being provided while training. DeepSNP results demonstrate the potential of DNN architectures to learn from genomic SNPa data and encourage further adaptation for CNV detection in SNPa and other genomic data types. [ABSTRACT FROM AUTHOR]