Multi-scale stacked sequential learning

Sequential learning is the discipline of machine learning that deals with dependent data such that neighboring labels exhibit some kind of relationship. The paper main contribution is two-fold: first, we generalize the stacked sequential learning, highlighting the key role of neighboring interactions modeling. Second, we propose an effective and efficient way of capturing and exploiting sequential correlations that takes into account long-range interactions. We tested the method on two tasks: text lines classification and image pixel classification. Results on these tasks clearly show that our approach outperforms the standard stacked sequential learning as well as state-of-the-art conditional random fields.     

Personalization and user verification in wearable systems using biometric walking patterns

In this article, a novel technique for user’s authentication and verification using gait as a biometric unobtrusive pattern is proposed. The method is based on a two stages pipeline. First, a general activity recognition classifier is personalized for an specific user using a small sample of her/his walking pattern. As a result, the system is much more selective with respect to the new walking pattern. A second stage verifies whether the user is an authorized one or not. This stage is defined as a one-class classification problem. In order to solve this problem, a four-layer architecture is built around the geometric concept of convex hull. This architecture allows to improve robustness to outliers, modeling non-convex shapes, and to take into account temporal coherence information. Two different scenarios are proposed as validation with two different wearable systems. First, a custom high-performance wearable system is built and used in a free environment. A second dataset is acquired from an Android-based commercial device in a ‘wild’ scenario with rough terrains, adversarial conditions, crowded places and obstacles. Results on both systems and datasets are very promising, reducing the verification error rates by an order of magnitude with respect to the state-of-the-art technologies.     

Emission cross sections and spectroscopy of Ho3+ laserchannels in KGd(WO4)2 single crystal

The spectroscopic properties of Ho³⁺ laser channels in KGd(WO₄)₂ crystals have been investigated using optical absorption, photoluminescence, and lifetime measurements. The radiative lifetimes of Ho³⁺ have been calculated through a Judd-Ofelt (JO) formalism using 300-K optical absorption results. The JO parameters obtained were Ω₂=15.35×10^-20 cm², Ω ₄=3.79×10^-20 cm², Ω₆ =1.69×10^-20 cm². The 7-300-K lifetimes obtained in diluted (8·10¹⁸ cm^-3) KGW:0.1% Ho samples are: τ(⁵F₃)≈0.9 μs, τ( ⁵S₂)=19-3.6 μs, and τ(⁵F₅ )≈1.1 μs. For Ho concentrations below 1.5×10²⁰ cm^-3, multiphonon emission is the main source of non radiative losses, and the temperature independent multiphonon probability in KGW is found to follow the energy gap law τ_ph ^-1(0)=βexp(-αΔE), where β=1.4×10^-7 s^-1, and α=1.4×10³ cm. Above this holmium concentration, energy transfer between Ho impurities also contributes to the losses. The spectral distributions of the Ho³⁺ emission cross section σ_EM for several laser channels are calculated in σ- and π-polarized configurations. The peak a σ_EM values achieved for transitions to the ⁵I₈ level are ≈2×10^-20 cm² in the σ-polarized configuration, and three main lasing peaks at 2.02, 2.05, and 2.07 μm are envisaged inside the ⁵I₇→⁵I₈ channel     

Fasting plasma glucose in screening for diabetes in the Taiwanese population

OBJECTIVE: To reveal the relationship between fasting and 2-h postload plasma glucose and to examine the appropriate fasting glucose cutoff as the primary screening test for diabetes. RESEARCH DESIGN AND METHODS: We recruited 5,303 subjects from preventive services of the National Cheng Kung University Hospital. Exclusion criteria were age <20 years, pregnancy, known diabetes, and a history of recent surgery, trauma, or illness. All subjects received the 75-g oral glucose tolerance test. The relationship between fasting and 2-h glucose was examined. Sensitivities, specificities, efficiency, and predictive values were assessed at different cutoffs of fasting glucose for prediction of diabetes. RESULTS: The best fit model for the relationship between fasting and 2-h glucose was fasting glucose = 4.914-0.060 x (2-h glucose) + 0.0144 x (2-h glucose)2. From this model, the fasting glucose was 6.0 mmol/l when 2-h glucose was 11.1 mmol/l. A fasting glucose with 6.25 mmol/l gave the same diabetes prevalence as the World Health Organization 2-h glucose criterion. When 7.8 mmol/l was the fasting glucose cutoff, the sensitivity was 28.5%. Lowering the cutoff from 7.8 to 7.0 mmol/l increased the sensitivity by 11.2% and slightly reduced the specificity and positive predictive value. If the cutoffs were 6.25 and 6.0 mmol/l, the sensitivity increased and the specificity and the positive predictive value decreased accordingly. CONCLUSIONS: Our results suggest that fasting glucose as a screening criterion for diabetes could be revised downward to 7.0 mmol/l, because the slight reduction of positive predictive value was more than balanced by an apparent increase of sensitivity and insignificant change of specificity.     

Dosimetric analysis of chimeric monoclonal antibody cMOv18 IgG in ovarian carcinoma patients after intraperitoneal and intravenous administration

In this study the potential of intraperitoneal (i.p.) and intravenous (i.v.) administration of chimeric iodine-131-labelled MOv18 IgG for radioimmunotherapy was determined. The dosimetry associated with both routes of administration of cMOv18 IgG was studied in patients. Eight patients suspected of having ovarian carcinoma received 150 MBq 131I-cMOv18 IgG i.p. Blood and urine were collected and serial gamma camera images were acquired. Another group of four patients received 7.5 MBq 131I-cMOv18 IgG i.v. For all patients, tissue biopsies were obtained at surgery. Activity in the blood after i.p. administration was described by a bi-exponential curve with a mean uptake and elimination half-life of 6.9+/-3.2 h and 160+/-45 h, respectively. For i.v. infusion the mean half-life for the elimination phase was 103+/-12 h. Cumulative excretion in the urine was 17%+/-3% ID and 21%+/-7% ID in 96 h for i.p. and i.v. administration, respectively. Scintigraphic images after i.p. administration showed accumulation in ovarian cancer lesions, while all other tissues showed decreasing activity with time. Tumour uptake determined in the ovarian cancer tissue specimens ranged from 3.4% to 12.3% ID/kg for i.p. administration and from 3.6% to 5.4% ID/kg for i.v. administration. Dosimetric analysis of the data indicated that 1.7-4.3 mGy/MBq and 1.7-2.2 mGy/MBq can be guided to solid or ascites cells after i.p. and i.v. administration, respectively. Assuming that an absorbed dose to the bone marrow of 2 Gy will be dose limiting, a total activity of 4.1 GBq 131I-cMOv18 IgG can be administered safely via the i.p. route and 3.5 GBq via the i.v. route. At this maximal tolerated dose, a maximum absorbed dose to 1-g tumours in the peritoneal cavity of 18 and 8 Gy can be reached after i.p. and i.v. administration, respectively. For the i. p. route of administration, dose estimates for the tumour are even higher when the electron dose of the peritoneal activity is also taken into account: total doses to the tumour of 30 Gy and 22 Gy will be absorbed at the tumour surface and at 0.2 mm depth, respectively. In conclusion, therapeutic tumour doses can be achieved with 131I-cMOv18 IgG in patients with intraperitoneal ovarian cancer lesions with no normal organ toxicity. The i.p. route of administration seems to be preferable to i.v. administration.     

V/Q mismatch without pulmonary emboli in children with histoplasmosis

In two children with histoplasmosis pulmonary perfusion and ventilation studies revealed mismatched abnormalities characterized by almost unilaterally absent perfusion but normal ventilation in the right lung. Chest roentgenograms demonstrated right hilar enlargement and pulmonary contrast angiograms revealed narrowing of the right pulmonary arteries by extrinsic fibrotic granulomata but no pulmonary emboli were present. Fibrosing mediastinitis due to histoplasmosis was found by mediastinoscopic examination in one of the patients. Non-embolic causes of V/Q mismatch lung imaging are discussed briefly, and the correlation with clinical findings is stressed for the diagnosis of pulmonary emboli.