Machine learning classification of surgical pathology reports and chunk recognition for information extraction noise reduction

Background and aims: Machine learning techniques for the text mining of cancer-related clinical documents have not been sufficiently explored. Here some techniques are presented for the pre-processing of free-text breast cancer pathology reports, with the aim of facilitating the extraction of information relevant to cancer staging.

Materials and methods: The first technique was implemented using the freely available software RapidMiner to classify the reports according to their general layout: ‘semi-structured’ and ‘unstructured’. The second technique was developed using the open source language engineering framework GATE and aimed at the prediction of chunks of the report text containing information pertaining to the cancer morphology, the tumour size, its hormone receptor status and the number of positive nodes. The classifiers were trained and tested respectively on sets of 635 and 163 manually classified or annotated reports, from the Northern Ireland Cancer Registry.

Results: The best result of 99.4% accuracy – which included only one semi-structured report predicted as unstructured – was produced by the layout classifier with the k nearest algorithm, using the binary term occurrence word vector type with stopword filter and pruning. For chunk recognition, the best results were found using the PAUM algorithm with the same parameters for all cases, except for the prediction of chunks containing cancer morphology. For semi-structured reports the performance ranged from 0.97 to 0.94 and from 0.92 to 0.83 in precision and recall, while for unstructured reports performance ranged from 0.91 to 0.64 and from 0.68 to 0.41 in precision and recall. Poor results were found when the classifier was trained on semi-structured reports but tested on unstructured.

Conclusions: These results show that it is possible and beneficial to predict the layout of reports and that the accuracy of prediction of which segments of a report may contain certain information is sensitive to the report layout and the type of information sought.

Oxygen targeting in preterm infants using the Masimo SET Radical pulse oximeter

BACKGROUND: A pretrial clinical improvement project for the BOOST-II UK trial of oxygen saturation targeting revealed an artefact affecting saturation profiles obtained from the Masimo Set Radical pulse oximeter.

METHODS: Saturation was recorded every 10 s for up to 2 weeks in 176 oxygen dependent preterm infants in 35 UK and Irish neonatal units between August 2006 and April 2009 using Masimo SET Radical pulse oximeters. Frequency distributions of % time at each saturation were plotted. An artefact affecting the saturation distribution was found to be attributable to the oximeter's internal calibration algorithm. Revised software was installed and saturation distributions obtained were compared with four other current oximeters in paired studies.

RESULTS: There was a reduction in saturation values of 87-90%. Values above 87% were elevated by up to 2%, giving a relative excess of higher values. The software revision eliminated this, improving the distribution of saturation values. In paired comparisons with four current commercially available oximeters, Masimo oximeters with the revised software returned similar saturation distributions.

CONCLUSIONS: A characteristic of the software algorithm reduces the frequency of saturations of 87-90% and increases the frequency of higher values returned by the Masimo SET Radical pulse oximeter. This effect, which remains within the recommended standards for accuracy, is removed by installing revised software (board firmware V4.8 or higher). Because this observation is likely to influence oxygen targeting, it should be considered in the analysis of the oxygen trial results to maximise their generalisability.

Understanding catalytic reactions over zeolites: A density functional theory study of selective catalytic reduction of NOX by NH3 over Cu-SAPO-34

Metal exchanged CHA-type (SAPO-34 and SSZ-13) zeolites are promising catalysts for selective catalytic reduction (SCR) of NOx by NH3. However, the understanding of the process at the molecular level is still limited, which hinders the identification of its mechanism and the design of more efficient zeolite catalysts. In this work, modelling the reaction over Cu-SAPO-34, a periodic density functional theory (DFT) study of NH3-SCR was performed using hybrid functional with the consideration of van der Waals (vdW) interactions. A mechanism with a low N–N coupling barrier is proposed to account for the activation of NO. The redox cycle of Cu2+ and Cu+, which is crucial for the SCR process, is identified with detailed analyses. Besides, the decomposition of NH2NO is shown to readily occur on the Brønsted acid site by a hydrogen push-pull mechanism, confirming the collective efforts of Brønsted acid and Lewis acid (Cu2+) sites. The special electronic and structural properties of Cu-SAPO-34 are demonstrated to play an essential role the reaction, which may have a general implication on the understanding of zeolite catalysis.