Effect of image resolution on automated classification of chest X-rays.

Purpose: Deep learning (DL) models have received much attention lately for their ability to achieve expert-level performance on the accurate automated analysis of chest X-rays (CXRs). Recently available public CXR datasets include high resolution images, but state-of-the-art models are trained on reduced size images due to limitations on graphics processing unit memory and training time. As computing hardware continues to advance, it has become feasible to train deep convolutional neural networks on high-resolution images without sacrificing detail by downscaling. This study examines the effect of increased resolution on CXR classification performance. Approach: We used the publicly available MIMIC-CXR-JPG dataset, comprising 377,110 high resolution CXR images for this study. We applied image downscaling from native resolution to 2048×2048 pixels, 1024×1024 pixels, 512×512 pixels, and 256×256 pixels and then we used the DenseNet121 and EfficientNet-B4 DL models to evaluate clinical task performance using these four downscaled image resolutions. Results: We find that while some clinical findings are more reliably labeled using high resolutions, many other findings are actually labeled better using downscaled inputs. We qualitatively verify that tasks requiring a large receptive field are better suited to downscaled low resolution input images, by inspecting effective receptive fields and class activation maps of trained models. Finally, we show that stacking an ensemble across resolutions outperforms each individual learner at all input resolutions while providing interpretable scale weights, indicating that diverse information is extracted across resolutions. Conclusions: This study suggests that instead of focusing solely on the finest image resolutions, multi-scale features should be emphasized for information extraction from high-resolution CXRs.

Study of the Etiological Factors of Failures of Myringoplasty: A Observational Study.

This Study entitled "the study of the aetiological factors of failures of myringoplasty" was planned to find the various aetiological factors causing failure of myringoplasty and suggests remedial steps to improve the results in future. This observational study was carried in the Department of E.N.T, on patients who had undergone myringoplasty from Jan 2013 to June 2016. Total 905 patients operated in the above mentioned duration. 685 patients were operated during Jan 2013 to June 2015, who were studied retrospectively (only 540 came for follow up), and 220 patients were operated during June 2015 to June 2016 were studied prospectively. The patients who came to the OPD for follow up, were thoroughly reviewed and the re-perforation was seen in 53 patients. Total nine factors causing myringoplasty failure were studied, 4 factors i.e. size of perforation, site of perforation, eustachian tube function, and surgeons experience significantly affects the surgical outcome of myringoplasty while age, tympanosclerotic patch, type of anaesthesia, wet or dry graft placement, and deviated nasal septum does not affects the myringoplasty failure. For improvement of result in future, Eustachian tube dysfunction, symptomatic deviated nasal septum should be corrected before the surgery. Anterior and larger perforation should be operated by experienced surgeon.

Improved Optical Multiplexing with Temporal DNA Barcodes.

Many biochemical events of importance are complex and dynamic. Fluorescence microscopy offers a versatile solution to study the dynamics of biology at the mesoscale. An important challenge in the field is the simultaneous study of several objects of interest, referred to as optical multiplexing. For improved multiplexing, some prior techniques used repeated reporter washing or the geometry of nanostructures; however, these techniques require complex nanostructure assembly, multiple reporters, or advanced multistep drift correction. Here we propose a time-based approach, for improved optical multiplexing, that uses readily available inexpensive reporters and requires minimal preparation efforts. We program short DNA strands, referred hereby as DNA devices, such that they undergo unique conformation changes in the presence of the dye-labeled reporters. The universal fluorescent reporter transiently binds with the devices to report their activity. Since each device is programmed to exhibit different hybridization kinetics, their fluorescent time trace, referred to as the temporal barcode, will be unique. We model our devices using continuous-time Markov chains and use stochastic simulation algorithm to generate their temporal patterns. We first ran simulation experiments with a small number of DNA devices, demonstrating several distinct temporal barcodes, all of which use a single dye color. Later, using nanostructure-based devices, we designed a much larger pool of temporal barcodes and used machine learning for classification of these barcodes. Our simulation experiments and design principles can aid in the experimental demonstration of the DNA devices.

Programming Temporal DNA Barcodes for Single-Molecule Fingerprinting.

Fluorescence microscopy enables simultaneous observation of the dynamics of single molecules in a large region of interest. Most traditional techniques employ either the geometry or the color of single molecules to uniquely identify (or barcode) different species of interest. However, these techniques require complex sample preparation and multicolor hardware setup. In this work, we introduce a time-based amplification-free single-molecule barcoding technique using easy-to-design nucleic acid strands. A dye-labeled complementary reporter strand transiently binds to the programmed nucleic acid strands to emit temporal intensity signals. We program the DNA strands to emit uniquely identifiable temporal signals for molecular-scale fingerprinting. Since the reporters bind transiently to DNA devices, our method offers relative immunity to photobleaching. We use a single universal reporter strand for all DNA devices making our design extremely cost-effective. We show DNA strands can be programmed for generating a multitude of uniquely identifiable molecular barcodes. Our technique can be easily incorporated with the existing orthogonal methods that use wavelength or geometry to generate a large pool of distinguishable molecular barcodes thereby enhancing the overall multiplexing capabilities of single-molecule imaging.