Postnatal gestational age estimation of newborns using Small Sample Deep Learning.

A baby's gestational age determines whether or not they are premature, which helps clinicians decide on suitable post-natal treatment. The most accurate dating methods use Ultrasound Scan (USS) machines, but these are expensive, require trained personnel and cannot always be deployed to remote areas. In the absence of USS, the Ballard Score, a postnatal clinical examination, can be used. However, this method is highly subjective and results vary widely depending on the experience of the examiner. Our main contribution is a novel system for automatic postnatal gestational age estimation using small sets of images of a newborn's face, foot and ear. Our two-stage architecture makes the most out of Convolutional Neural Networks trained on small sets of images to predict broad classes of gestational age, and then fuses the outputs of these discrete classes with a baby's weight to make fine-grained predictions of gestational age using Support Vector Regression. On a purpose-collected dataset of 130 babies, experiments show that our approach surpasses current automatic state-of-the-art postnatal methods and attains an expected error of 6 days. It is three times more accurate than the Ballard method. Making use of images improves predictions by 33% compared to using weight only. This indicates that even with a very small set of data, our method is a viable candidate for postnatal gestational age estimation in areas were USS is not available.

Identifying Chirality in Line Drawings of Molecules Using Imbalanced Dataset Sampler for a Multilabel Classification Task.

Chirality, the ability of some molecules to exist as two non-superimposable mirror images, profoundly influences both chemistry and biology. Advances in deep learning enable the automatic recognition of chemical structure diagrams, however, studies on discovering the molecule chirality are scarce and the machine-readable molecular representations are not always sufficient to fully support the encoding of this important property. Here, we pretrained networks on a ChEMBL+ dataset (79641 molecules) and fine-tuned them for the binary classification of chirality (achiral/chiral) or multilabel chirality type classifications (none/centre/axial/planar). To address the label combination imbalanced problem in the multilabel task, the study proposed a Formulated Imbalanced Dataset Sampler (FIDS) to sample a formulated amount of minority label combinations on top of the training set. On a 10-fold cross validation experiment using our CHIRAL dataset (1142 manually curated molecules), our models achieved up to an accuracy of 90 % in the binary task. In the multilabel task incorporated with FIDS, the overall performance increases from 87 % to 89 % and the accuracy per label combination can attained up to a 50 % increase. Through the study of heatmaps, our work also exemplified the potential of deep neural network to make predictions based on the actual location of chirality elements.

RevManHAL: towards automatic text generation in systematic reviews.

BACKGROUND: Systematic reviews are a key part of healthcare evaluation. They involve important painstaking but repetitive work. A major producer of systematic reviews, the Cochrane Collaboration, employs Review Manager (RevMan) programme-a software which assists reviewers and produces XML-structured files. This paper describes an add-on programme (RevManHAL) which helps auto-generate the abstract, results and discussion sections of RevMan-generated reviews in multiple languages. The paper also describes future developments for RevManHAL. METHODS: RevManHAL was created in Java using NetBeans by a programmer working full time for 2 months. RESULTS: The resulting open-source programme uses editable phrase banks to envelop text/numbers from within the prepared RevMan file in formatted readable text of a chosen language. In this way, considerable parts of the review's 'abstract', 'results' and 'discussion' sections are created and a phrase added to 'acknowledgements'. CONCLUSION: RevManHAL's output needs to be checked by reviewers, but already, from our experience within the Cochrane Schizophrenia Group (200 maintained reviews, 900 reviewers), RevManHAL has saved much time which is better employed thinking about the meaning of the data rather than restating them. Many more functions will become possible as review writing becomes increasingly automated.

AutoRoot: open-source software employing a novel image analysis approach to support fully-automated plant phenotyping.

BACKGROUND: Computer-based phenotyping of plants has risen in importance in recent years. Whilst much software has been written to aid phenotyping using image analysis, to date the vast majority has been only semi-automatic. However, such interaction is not desirable in high throughput approaches. Here, we present a system designed to analyse plant images in a completely automated manner, allowing genuine high throughput measurement of root traits. To do this we introduce a new set of proxy traits. RESULTS: We test the system on a new, automated image capture system, the Microphenotron, which is able to image many 1000s of roots/h. A simple experiment is presented, treating the plants with differing chemical conditions to produce different phenotypes. The automated imaging setup and the new software tool was used to measure proxy traits in each well. A correlation matrix was calculated across automated and manual measures, as a validation. Some particular proxy measures are very highly correlated with the manual measures (e.g. proxy length to manual length, r2 > 0.9). This suggests that while the automated measures are not directly equivalent to classic manual measures, they can be used to indicate phenotypic differences (hence the term, proxy). In addition, the raw discriminative power of the new proxy traits was examined. Principal component analysis was calculated across all proxy measures over two phenotypically-different groups of plants. Many of the proxy traits can be used to separate the data in the two conditions. CONCLUSION: The new proxy traits proposed tend to correlate well with equivalent manual measures, where these exist. Additionally, the new measures display strong discriminative power. It is suggested that for particular phenotypic differences, different traits will be relevant, and not all will have meaningful manual equivalent measures. However, approaches such as PCA can be used to interrogate the resulting data to identify differences between datasets. Select images can then be carefully manually inspected if the nature of the precise differences is required. We suggest such flexible measurement approaches are necessary for fully automated, high throughput systems such as the Microphenotron.