Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.

Resistive-Polymer Versus Forced-Air Warming: Comparable Efficacy in Orthopedic Patients

BACKGROUND: Several adverse consequences are caused by mild perioperative hypothermia. Maintaining normothermia with patient warming systems, today mostly with forced air (FA), has thus become a standard procedure during anesthesia. Recently, a polymer-based resistive patient warming system was developed. We compared the efficacy of a widely distributed FA system with the resistive-polymer (RP) system in a prospective, randomized clinical study. METHODS: Eighty patients scheduled for orthopedic surgery were randomized to either FA warming (Bair Hugger warming blanket #522 and blower #750, Arizant, Eden Prairie, MN) or RP warming (Hot Dog Multi-Position Blanket and Hot Dog controller, Augustine Biomedical, Eden Prairie, MN). Core temperature, skin temperature (head, upper and lower arm, chest, abdomen, back, thigh, and calf), and room temperature (general and near the patient) were recorded continuously. RESULTS: After an initial decrease, core temperatures increased in both groups at comparable rates (FA: 0.33 degrees C/h +/- 0.34 degrees C/h; RP: 0.29 degrees C/h +/- 0.35 degrees C/h; P = 0.6). There was also no difference in the course of mean skin and mean body (core) temperature. FA warming increased the environment close to the patient (the workplace of anesthesiologists and surgeons) more than RP warming (24.4 degrees C +/- 5.2 degrees C for FA vs 22.6 degrees C +/- 1.9 degrees C for RP at 30 minutes; P(AUC) <0.01). CONCLUSION: RP warming performed as efficiently as FA warming in patients undergoing orthopedic surgery.

Numerical model of the human cornea based on stromal microstructure: identification of mechanical properties for two age groups

The optical quality of the human eye mainly depends on the refractive performance of the cornea. The shape of the cornea is a mechanical balance between intraocular pressure and tissue intrinsic stiffness. Several surgical procedures in ophthalmology alter the biomechanics of the cornea to provoke local or global curvature changes for vision correction. Legitimated by the large number of surgical interventions performed every day, the demand for a deeper understanding of corneal biomechanics is rising to improve the safety of procedures and medical devices. The aim of our work is to propose a numerical model of corneal biomechanics, based on the stromal microstructure. Our novel anisotropic constitutive material law features a probabilistic weighting approach to model collagen fiber distribution as observed on human cornea by Xray scattering analysis (Aghamohammadzadeh et. al., Structure, February 2004). Furthermore, collagen cross-linking was explicitly included in the strain energy function. Results showed that the proposed model is able to successfully reproduce both inflation and extensiometry experimental data (Elsheikh et. al., Curr Eye Res, 2007; Elsheikh et. al., Exp Eye Res, May 2008). In addition, the mechanical properties calculated for patients of different age groups (Group A: 65-79 years; Group B: 80-95 years) demonstrate an increased collagen cross-linking, and a decrease in collagen fiber elasticity from younger to older specimen. These findings correspond to what is known about maturing fibrous biological tissue. Since the presented model can handle different loading situations and includes the anisotropic distribution of collagen fibers, it has the potential to simulate clinical procedures involving nonsymmetrical tissue interventions. In the future, such mechanical model can be used to improve surgical planning and the design of next generation ophthalmic devices.

"Device landing zone" calcification, assessed by MSCT, as a predictive factor for pacemaker implantation after TAVI

After trans-catheter aortic valve implantation (TAVI), the need for postinterventional pacemaker (PM) implantation can occur in as many as 10-50% of cases, but it is not yet clear, how this need can be predicted. The aim of this study was to assess the possible predictive factors of post TAVI PM implantation based on Computed Tomography (CT) measured aortic valve calcification and its distribution.

Forced-use therapy for children with cerebral palsy in the community setting: a single-blinded randomized controlled pilot trial

The aim of this study was to elucidate the feasibility, efficacy, and sustainability of a home-based, two-week, forced-use therapy (FUT) program for children with hemiplegic cerebral palsy (CP).

Assessment of bronchodilator responsiveness in preschool children using forced oscillations

BACKGROUND: The forced oscillation technique (FOT) requires minimal patient cooperation and is feasible in preschool children. Few data exist on respiratory function changes measured using FOT following inhaled bronchodilators (BD) in healthy young children, limiting the clinical applications of BD testing in this age group. A study was undertaken to determine the most appropriate method of quantifying BD responses using FOT in healthy young children and those with common respiratory conditions including cystic fibrosis, neonatal chronic lung disease and asthma and/or current wheeze. METHODS: A pseudorandom FOT signal (4-48 Hz) was used to examine respiratory resistance and reactance at 6, 8 and 10 Hz; 3-5 acceptable measurements were made before and 15 min after the administration of salbutamol. The post-BD response was expressed in absolute and relative (percentage of baseline) terms. RESULTS: Significant BD responses were seen in all groups. Absolute changes in BD responses were related to baseline lung function within each group. Relative changes in BD responses were less dependent on baseline lung function and were independent of height in healthy children. Those with neonatal chronic lung disease showed a strong baseline dependence in their responses. The BD response in children with cystic fibrosis, asthma or wheeze (based on both group mean data and number of responders) was not greater than in healthy children. CONCLUSIONS: The BD response assessed by the FOT in preschool children should be expressed as a relative change to account for the effect of baseline lung function. The limits for a positive BD response of -40% and 65% for respiratory resistance and reactance, respectively, are recommended.

Applied Graph Theory in Computer Vision and Pattern Recognition

This book will serve as a foundation for a variety of useful applications of graph theory to computer vision, pattern recognition, and related areas. It covers a representative set of novel graph-theoretic methods for complex computer vision and pattern recognition tasks. The first part of the book presents the application of graph theory to low-level processing of digital images such as a new method for partitioning a given image into a hierarchy of homogeneous areas using graph pyramids, or a study of the relationship between graph theory and digital topology. Part II presents graph-theoretic learning algorithms for high-level computer vision and pattern recognition applications, including a survey of graph based methodologies for pattern recognition and computer vision, a presentation of a series of computationally efficient algorithms for testing graph isomorphism and related graph matching tasks in pattern recognition and a new graph distance measure to be used for solving graph matching problems. Finally, Part III provides detailed descriptions of several applications of graph-based methods to real-world pattern recognition tasks. It includes a critical review of the main graph-based and structural methods for fingerprint classification, a new method to visualize time series of graphs, and potential applications in computer network monitoring and abnormal event detection.

Resistive polymer versus forced-air warming: comparable heat transfer and core rewarming rates in volunteers

BACKGROUND: Mild perioperative hypothermia increases the risk of several severe complications. Perioperative patient warming to preserve normothermia has thus become routine, with forced-air warming being used most often. In previous studies, various resistive warming systems have shown mixed results in comparison with forced-air. Recently, a polymer-based resistive patient warming system has been developed. We compared the efficacy of a standard forced-air warming system with the resistive polymer system in volunteers. METHODS: Eight healthy volunteers participated, each on two separate study days. Unanesthetized volunteers were cooled to a core temperature (tympanic membrane) of 34 degrees C by application of forced-air at 10 degrees C and a circulating-water mattress at 4 degrees C. Meperidine and buspirone were administered to prevent shivering. In a randomly designated order, volunteers were then rewarmed (until their core temperatures reached 36 degrees C) with one of the following active warming systems: (1) forced-air warming (Bair Hugger warming cover #300, blower #750, Arizant, Eden Prairie, MN); or (2) polymer fiber resistive warming (HotDog whole body blanket, HotDog standard controller, Augustine Biomedical, Eden Prairie, MN). The alternate system was used on the second study day. Metabolic heat production, cutaneous heat loss, and core temperature were measured. RESULTS: Metabolic heat production and cutaneous heat loss were similar with each system. After a 30-min delay, core temperature increased nearly linearly by 0.98 (95% confidence interval 0.91-1.04) degrees C/h with forced-air and by 0.92 (0.85-1.00) degrees C/h with resistive heating (P = 0.4). CONCLUSIONS: Heating efficacy and core rewarming rates were similar with full-body forced-air and full-body resistive polymer heating in healthy volunteers.

Microscopic wire guide-based orotracheal mouse intubation: description, evaluation and comparison with transillumination

Airway access is needed for a number of experimental animal models, and the majority of animal research is based on mouse models. Anatomical conditions in mice are small, and the narrow glottic opening allows intubation only with a subtle technique. We therefore developed a microscopic endotracheal intubation method with a wire guide technique in mice anaesthetized with halothane in oxygen. The mouse is hung perpendicularly with its incisors on a thread fixed on a vertical plate. The tongue is placed with a pair of forceps between the left hand's thumb and forefinger and slightly pulled, while the neck and thorax are positioned using the third and fourth fingers. By doing so, the neck can be slightly stretched, which allows optimal visualization of the larynx and the vocal cords. To ensure a safe intubation, a fine wire guide is placed under vision between the vocal cords and advanced about 5 mm into the trachea. An intravenous 22G x 1 in. plastic or Teflon catheter is guided over this wire. In a series of 41 mice, between 21 and 38 g, the success rate for the first intubation attempt was >95%. Certainty of the judgement procedure was 100% and success rate was higher using the described method when compared with a transillumination method in a further series. The technique is safe, less invasive than tracheostomy and suitable for controlled ventilation and pulmonary substance application.

Method of collaborative detection of autonomous transport vehicles based on laser rangefinder data

To master changing performance demands, autonomous transport vehicles are deployed to make inhouse material flow applications more flexible. The socalled cellular transport system consists of a multitude of small scale transport vehicles which shall be able to form a swarm. Therefore the vehicles need to detect each other, exchange information amongst each other and sense their environment. By provision of peripherally acquired information of other transport entities, more convenient decisions can be made in terms of navigation and collision avoidance. This paper is a contribution to collective utilization of sensor data in the swarm of cellular transport vehicles.

Assessing dynamics of forced livestock movements, livelihoods and future development options for pastoralists/agro-pastoralists in Ruvuma and Lindi Regions, in the Southern Tanzania

The study that aimed at understanding the dynamics of forced livestock movements and pastoral livelihood and development options was conducted in Lindi and Ruvuma regions, using both formal and informal approaches. Data were collected from 60 randomly selected Agro-pastoralists/Pastoralists and native farmers using a structured questionnaire. Four villages were involved; two in Lindi region (Matandu and Mkwajuni) and the other two in Ruvuma region (Gumbiro and Muhuwesi). Data were analyzed using descriptive statistics of SPSS to generate means and frequencies. The results indicate that a large number of animals moved into the study area following the eviction order of the government in Ihefu wetlands in 2006/2007. Lindi region was earmarked by the government to receive all the evicted pastoralists. However, by 2008 only 30% of the total cattle that were expected to move into the region had been received. Deaths of many animals on transit, selling of the animals to pay for transportation and other costs while on transit and many pastoralists settling in Coastal and Ruvuma regions before reaching their destinations were reported to be the reasons for the discrepancy observed. To mitigate anticipated conflicts between farmers and pastoralists, Participatory Land Use Management (PLUM) plans were developed in all the study villages in order to demarcate village land area into different uses, including grazing, cropping, settlement and forests. Land units for grazing were supposed to be provided with all necessary livestock infrastructures (dips, charcoal dams, livestock markets and stock routes). However, the land use plans were not able to prevent the anticipated conflicts because most of the livestock infrastructures were lacking, the land use boundaries were not clearly demarcated and there was limited enforcement of village by-laws, since most had not been enacted by the respective district councils. Similarly, the areas allocated for grazing were inadequate for the number of livestock available and thus the carrying capacity exceeded. Thus, land resource-based conflicts between farmers and pastoralists were emerging in the study areas for the reason that most of the important components in the PLUM plans were not in place. Nevertheless, the arrival of pastoralists in the study areas had positive effects on food security and growth of social interactions between pastoralists and farmers including marriages between them. Environmental degradations due to the arrival of livestock were also not evident. Thus, there is a need for the government to purposely set aside enough grazing land with all necessary infrastructures in place for the agro-pastoral/pastoral communities in the country.

Towards an SMS-based Social Network for Health Workers in Rural Areas in Myanmar

Seventy percent of the population in Myanmar lives in rural areas. Although health workers are adequately trained, they are overburdened due to understaffing and insufficient supplies. Literature confirms that information and communication technologies can extend the reach of healthcare. In this paper, we present an SMS-based social network that aims to help health workers to interact with other medical professionals through topic-based message delivery. Topics describe interests of users and the content of message. A message is delivered by matching message content with user interests. Users describe topics as ICD- 10 codes, a comprehensive medical taxonomy. In this ICD-10 coded SMS, a set of prearranged codes provides a common language for users to send structured information that fits inside an SMS.

Graphical Model-Based Vertebra Identification from X-Ray Image(s)

Automated identification of vertebrae from X-ray image(s) is an important step for various medical image computing tasks such as 2D/3D rigid and non-rigid registration. In this chapter we present a graphical model-based solution for automated vertebra identification from X-ray image(s). Our solution does not ask for a training process using training data and has the capability to automatically determine the number of vertebrae visible in the image(s). This is achieved by combining a graphical model-based maximum a posterior probability (MAP) estimate with a mean-shift based clustering. Experiments conducted on simulated X-ray images as well as on a low-dose low quality X-ray spinal image of a scoliotic patient verified its performance.