S.R. Amendolia

A comparative study of K-Nearest Neighbour, Support Vector Machine and Multi-Layer Perceptron for Thalassemia screening

In this paper, we investigate the feasibility of two typical techniques of Pattern Recognition in the classification for Thalassemia screening. They are the Support Vector Machine (SVM) and the K-Nearest Neighbour (KNN). We compare SVM and KNN with a Multi-Layer Perceptron (MLP) classifier. We propose a two-classifier system based on SVM. The first layer is used to differentiate between pathological and non-pathological cases while the second layer is used to discriminate between two different pathologies (α-thalassemia carrier against β-thalassemia carrier) from the first output layer (pathological cases). Using the parameters sensitivity (percentage of pathologic cases correctly classified) and specificity (percentage of non-pathologic cases correctly classified), the results obtained with this analysis show that the MLP classifier gives slightly better results than SVM although the amount of data available is limited. Both techniques enable thalassemia carriers to be discriminated from healthy subjects with 95% specificity, although the sensitivity of MLP is 92% while that of SVM is 83%.

The AMchip: a full-custom CMOS VLSI associative memory for pattern recognition

An associative memory full custom CMOS VLSI chip (AMchip), to be used in fast trigger systems for pattern recognition has been designed and is being successfully tested at INFN in Pisa. The AMchip is the first full-custom associative memory IC developed for high-energy physics. It contains about 140000 MOS transistors, has been realized in 1.5- mu m, double-metal, silicon gate CMOS technology, and is housed in a 120-pin package. The AMchip has been designed to be used with any kind of detector which provides output in hits coordinates, such as, for example, a silicon microstrip detector. The authors plan to realize a novel AMchip version using submicron technology and new circuit solutions, improving the pattern capacity by a factor 4, and significantly improving the speed. These versions will be developed to match new high-energy physics experiments specific requirements (for example, those of the Collision Detector Facility Silicon Vertex Tracker).<<ETX>>

Low contrast imaging with a GaAs pixel digital detector

A digital mammography system based on a GaAs pixel detector has been developed by the INFN (Istituto Nazionale di Fisica Nucleare) collaboration MED46. The high atomic number makes the GaAs a very efficient material for low energy X-ray detection (10-30 keV is the typical energy range used in mammography). Low contrast details can be detected with a significant dose reduction to the patient. The system presented in this paper consists of a 4096 pixel matrix built on a 200 /spl mu/m thick semi-insulating GaAs substrate. The pixel size is 170/spl times/170 /spl mu/m/sup 2/ for a total active area of 1.18 cm/sup 2/. The detector is bump-bonded to a VLSI front-end chip which implements a single-photon counting architecture. This feature allows to enhance the radiographic contrast detection with respect to charge integrating devices. The system has been tested by using a standard mammographic tube. Images of mammographic phantoms will be presented and compared with radiographs obtained with traditional film/screen systems. Monte Carlo simulations have been also performed to evaluate the imaging capability of the system. Comparison with simulations and experimental results will be shown.

Use of silicon and GaAs pixel detectors for digital autoradiography

Solid state detectors made of Si (4.8/spl times/8 mm/sup 2/) and GaAs (6.4/spl times/8 mm/sup 2/) pixel matrices bump-bonded to the Omega2 and Omega3 electronic read-out systems, developed at CERN for H.E.P. experiments, have been used to obtain autoradiographic images of clusters of human epithelial cells and DNA fragments separated via electrophoresis, both labeled with /sup 32/P. The system has shown a good minimum detectable activity per unit area of 2.10/sup -4/ cps mm/sup -2/, and has proved linear for a count rate in the range 0.2-20 cpa, typical of autoradiography. The pixel dimensions are 75/spl times/500 /spl mu/m/sup 2/ (Si-Omega2) and 50/spl times/500 /spl mu/m/sup 2/ (GaAs-Omega3), respectively. We are able to clearly localize clusters of cells which have incorporated the radioactive tracer and DNA fragments on an electrophoretic gel on paper (blots).

AUTORADIOGRAPHY WITH SILICON STRIP DETECTORS

A digital autoradiography system based on double sided silicon strip detectors (1.6 × 1.6 mm2 active surface with 100 μm pitch) has been developed and successfully tested with beta-emitting tracers. It is shown here that the system is able to perform imaging of organic material with specific sensitivity as small as 0.002 nCi/mm2, and to record activity measurements with good linearity in the range 0.002–20 nCi/mm2. Autoradiographic images of clusters of mammary cells marked with ortho-(32P)phosphate, obtained with an exposure time of about 10 min are presented.

A Real-Time Classification System of Thalassemic Pathologies Based on Artificial Neural Networks

Thalassemias are pathologies that derive from genetic defects of the globin genes. The most common defects among the population affect the genes that are involved in the synthesis of α and β chains. The main aspects of these pathologies are well explained from a biochemical and genetic point of view. The diagnosis is fundamentally based on hematologic and genetic tests. A genetic analysis is particularly important to determine the carriers of α-thalassemia, whose identification by means of the hematologic parameters is more difficult in comparison with heterozygotes for β-thalassemia. This work investigates the use of artificial neural networks (ANNs) for the classification of thalassemic pathologies using the hematologic parameters resulting from hemochromocytometric analysis only. Different combinations of ANNs are reported, which allow thalassemia carriers to be discriminated from normals with 94% classification accuracy, 92% sensitivity, and 95% specificity. On the basis of these results, an automated system that allows real-time support for diagnoses is proposed. The automated system interfaces a hemochromo analyzer to a simple PC.

The AMchip: a VLSI associative memory for track finding

Abstract An associative memory to be used for super-fast track finding in future high energy physics experiments, has been implemented on silicon as a full-custom CMOS VLSI chip (the AMchip). The first prototype has been designed and successfully tested at INFN in Pisa. It is implemented in 1.6 μm, double metal, silicon gate CMOS technology and contains about 140 000 MOS transistors on a 1 × 1 cm 2 silicon chip.

Experimental study of LEC GaAs detectors for X-ray digital radiography

Abstract In previous studies, various semi-insulating LEC GaAs crystals were irradiated with photons in the diagnostic energy range (20–100 keV), in view of a possible application in digital radiography. Solid-state and irradiation measurements, together with Monte Carlo simulations, have indicated good candidates for this application among the crystals we have investigated. In this paper we present results concerning the detection characteristics (detection efficiency, charge-collection efficiency and energy resolution as functions of the bias voltage) of one of these materials and the images obtained by a pixel detector made on the same material and a bump-bonded electronic system.

Semiconductor pixel detectors for digital mammography

Abstract We present some results obtained with silicon and gallium arsenide pixel detectors to be applied in the field of digital mammography. Even though GaAs is suitable for medical imaging applications thanks to its atomic number, which allows a very good detection efficiency, it often contains an high concentrations of traps which decrease the charge collection efficiency (CCE). So we have analysed both electrical and spectroscopic performance of different SI GaAs diodes as a function of concentrations of dopants in the substrate, in order to find a material by which we can obtain a CCE allowing the detection of all the photons that interact in the detector. Nevertheless to be able to detect low contrast details, efficiency and CCE are not the only parameters to be optimized; also the stability of the detection system is fundamental. In the past we have worked with Si pixel detectors; even if its atomic number does not allow a good detection efficiency at standard thickness, it has a very high stability. So keeping in mind the need to increase the Silicon detection efficiency we performed simulations to study the behaviour of the electrical potential in order to find a geometry to avoid the risk of electrical breakdown.

GaAs detector optimization for different medical imaging applications

Abstract We have investigated the detection performance of GaAs detectors made with different thickness and contact geometries. A comparison is made between these detection capabilities and the imaging requirements for the following medical applications: digital mammography, digital chest radiography and nuclear medicine. Experimental results and preliminary images are presented and discussed.