Thanks to their capability to learn generalizable descriptors directly from images, deep Convolutional Neural Networks (CNNs) seem the ideal solution to most pattern recognition problems. On the other hand, to learn the image representation, CNNs need huge sets of annotated samples that are unfeasible in many every-day scenarios. This is the case, for example, of Computer-Aided Diagnosis (CAD) systems for digital pathology, where additional challenges are posed by the high variability of the cancerous tissue characteristics. In our experiments, state-of-the-art CNNs trained from scratch on histological images were less accurate and less robust to variability than a traditional machine learning framework, highlighting all the issues of fully training deep networks with limited data from real patients. To solve this problem, we designed and compared three transfer learning frameworks, leveraging CNNs pre-trained on non-medical images. This approach obtained very high accuracy, requiring much less computational resource for the training. Our findings demonstrate that transfer learning is a solution to the automated classification of histological samples and solves the problem of designing accurate and computationally-efficient CAD systems with limited training data.

Despite the many advantages and increasing adoption of Electron Beam Powder Bed Fusion (PBF-EB) additive manufacturing by industry, current PBF-EB systems remain largely unstable and prone to unpredictable anomalous behaviours. Additionally, although featuring in-situ process monitoring, PBF-EB systems show limited capabilities in terms of timely identification of process failures, which may result into considerable wastage of production time and materials. These aspects are commonly recognized as barriers for the industrial breakthrough of PBF-EB technologies. On top of these considerations, in our research we aim at introducing real-time anomaly detection capabilities into the PBF-EB process. To do so, we build our case-study on top of a Arcam EBM A2X system, one of the most diffused PBF-EB machines in industry, and make access to the most relevant variables made available by this machine during the layering process. Thus, seeking a proficient interpretation of such data, we introduce a deep learning autoencoder-based anomaly detection framework. We demonstrate that this framework is able not only to early identify anomalous patterns from such data in real-time during the process with a F1 score around 90%, but also to anticipate the failure of the current job by 6 h, on average, and in one case by almost 20 h. This avoids waste of production time and opens the way to a more controllable PBF-EB process.

The prognosis of renal cell carcinoma (RCC) malignant neoplasms deeply relies on an accurate determination of the histological subtype, which currently involves the light microscopy visual analysis of histological slides, considering notably tumor architecture and cytology. RCC subtyping is therefore a time-consuming and tedious process, sometimes requiring expert review, with great impact on diagnosis, prognosis and treatment of RCC neoplasms. In this study, we investigate the automatic RCC subtyping classification of 91 patients, diagnosed with clear cell RCC, papillary RCC, chromophobe RCC, or renal oncocytoma, through deep learning based methodologies. We show how the classification performance of several state-of-the-art Convolutional Neural Networks (CNNs) are perfectible among the different RCC subtypes. Thus, we introduce a new classification model leveraging a combination of supervised deep learning models (specifically CNNs) and pathologist’s expertise, giving birth to a hybrid approach that we termed ExpertDeepTree (ExpertDT). Our findings prove ExpertDT’s superior capability in the RCC subtyping task, with respect to traditional CNNs, and suggest that introducing some expert-based knowledge into deep learning models may be a valuable solution for complex classification cases.

So far, very little attention has been paid to the role of the autonomic nervous system in augmentative alternative communication solutions. In this regard, the pupil near reflex, one component of the triadic accommodative response to a visual plane shift in-depth, may play a key role. Such reflex does not necessitate any requirement of skeletal muscles, and thus may be preserved in diseases affecting somatic motoneurons, such as the amyotrophic lateral sclerosis. On this basis, the pupillary accommodative response, i.e. the pupil constriction in response to a far-to-near visual shift in depth, can constitute an effective communication tool, bypassing the well-known limitations of canonical eye-trackers, which require recruitment of skeletal muscles. This paper introduces e-Pupil: a low-cost, stand-alone, portable and smart device exploiting the pupillary accommodative response as a communication tool. The Internet of Things plays a crucial role in our device, guaranteeing portability, accessibility and usability, as well as several remote functionalities. We propose two different routes to communicate with the external world, both built upon the identification of the pupil constriction event in response to a voluntary far-to-near visual shift. The experimental validation proves the reliability of the system as well as its intrinsic simplicity of use.