Abstract
Evaluation of lung function is central to the management of chronic obstructive respiratory diseases. It is typically evaluated with a spirometer by a specialized health professional, who ensures the correct execution of a forced expiratory manoeuvre (FEM). Audio recording of a FEM using a smart device embedded microphone can be used to self-monitor lung function between clinical visits. The challenge of microphone spirometry is to ensure the validity and reliability of the FEM, in the absence of a health professional. In particular, the absence of a mouthpiece may allow excessive mouth closure, leading to an incorrect manoeuvre. In this work, a strategy to automatically assess the correct execution of the FEM is proposed and validated. Using 498 FEM recordings, both specificity and sensitivity attained were above 90%. This method provides immediate feedback to the user, by grading the manoeuvre in a visual scale, promoting the repetition of the FEM when needed. © 2020, Springer Nature Switzerland AG.

Abstract

Abstract

Abstract
The small amount of public available medical images hinders the use of deep learning techniques for mammogram automatic diagnosis. Deep learning methods require large annotated training sets to be effective, however medical datasets are costly to obtain and suffer from large variability. In this work, a lightweight deep learning pipeline to detect, segment and classify anomalies in mammogram images is presented. First, data augmentation using the ground-truth annotation is performed and used by a cascade segmentation and classification methods. Results are obtained using the INbreast public database in the context of lesion detection and BI-RADS classification. Moreover, a pre-trained Convolutional Neural Network using ResNet50 is modified to generate the lesion regions proposals followed by a false positive reduction and contour refinement stages while a pre-trained VGG16 network is fine-tuned to classify mammograms. The detection and segmentation stage results show that the cascade configuration achieves a DICE of 0.83 without massive training while the multi-class classification exhibits an MAE of 0.58 with data augmentation. © Springer Nature Switzerland AG 2019.

Abstract
The Sternum is a human bone located in the anterior area of the thoracic cage. It is present in most of the axial cuts provided from the Magnetic Resonance Imaging (MRI) acquisitions, used in the medical field. Detecting the Sternum is relevant as it contains rigid key-points for 3D model reconstructions, assisting in the planning and evaluation of several surgical procedures, and for atlas development by segmenting structures in anatomical proximity. In the absence of applicable approaches for this specific problem, this paper focuses on two distinct automated methods for Sternum segmentation in MRI. The first, relies on K-Means (Clustering) to perform the segmentation, while the second encompasses the closed Minimum Path over the elliptical transformation of Gradient images. A dataset of 14 annotated acquisitions was used for evaluation. The results favored the Gradient approach over Clustering. © 2019 IEEE.