فصلنامه مهندسی پزشکی زیستی
سال یکم شماره 3 (پاییز 1386)

In this study, the need of a CAD system and its capabilities has been investigated and then a sample program for a mammographic CAD system proper to Iranian tropical patients was designed. In the first step, the analog mammographic images were digitized by 56 and 112 mm spatial resolution and then were processed by the designed sample program. Analysis and technical details for designing and implementing the program included for following steps: The capability of the program image displayer consisting of viewing four mammographic images from four breast views (RCC, RMLO, LCC, LMLO) in one window, determining breast region by background removing and other conventional preprocessing application tools; Software processing tools including theresholding, histogram, ROI determination; Patient information fields such as clinical information, conventional reporting section as used in radiological department in Iran; Computer-aided diagnostic section including proper diagnostic processing algorithm to automatic detection of breast abnormality. For instance the application of wavelet and fuzzy logic for detecting malignant clusters of microcalcification. The introduced mammographic CAD system can provide the collection, organizing and the availability of the patient local information. Therefore by using the prepared database the evaluation of the sensitivity and specifity of the detecting algorithm for comparison of different research methods would be possible.

In this paper, the role of Vector Quantizer Neural Network in classification of six types of ECG signals has been investigated using the features that extracted from Daubechies6 Wavelet transformation. The six types of signals are: normal beat, left bundle branch block beat, right bundle branch block beat, premature ventricular contraction paced beat and fusion of paced and normal beats. The required data were obtained from the MIT/BIH arrhythmia databases. By using the annotation files of the databases, the patterns of these six types of ECG signals were separated. Then, for better feature extraction, filtering and scaling on the patterns were applied. We used the energies of the last five detailed signals obtained from the exerting the Wavelet transformation in six levels, as the pattern features for Vector Quantizer Network training and testing. From each class, five hundred patterns were used for network training and one hundred patterns for testing. The results indicated %93.1 accuracy for six classes and above %94.3 for lesser than six classes. Then the rate of similarity and dissimilarity of the classes were considered. Finally, the results of this method were compared with some other methods in terms of accuracy.

A two dimensional finite element model for the human Haversian cortical bone is represented. The interstitial bone tissue, the osteons and the cement line were modeled as the matrix, the fibers and the interface, respectively. This was due to similarities between fiber-ceramic composite materials and the human Haversian cortical bone. The stress intensity factor in the microcrack tips vicinity was computed using the linear elastic fracture mechanics theory and assuming a plane strain condition. It was therefore possible to study the effect of microstructure and mechanical properties of Haversian cortical bone on microcrack propagation trajectory. The results indicated that this effect was limited to the vicinity of the osteon. If both osteon and cement line were assumed to be softer than the interstitial tissue, the stress intensity factor was increased when the crack distance to the osteon reduced. The stress intensity factor decreased if both osteon and cement line were assumed to be stiffer than the interstitial tissue. The resulting simulation indicated that the effect of existence of osteon on the stress intensity factor was no significance, if both the interstitial tissue and cement line were assumed either stiffer or softer than the osteon. Microcrack trajectory was observed to deviate from the osteon under tensile loading; indicating an independence from the mechanical properties of various tissues. In fact, the microcrack adopts a trajectory between the osteons, thereby increasing the necessary absorbed energy for fracture. This results in an increase in the human Haversian cortical bone toughness. The result of this finite element modeling has been confirmed by through evaluation and comparison made with experimental results.

A novel approach to surface electromyogram (sEMG) signal classification using its higher order statistics (HOS) is presented in this study. As the probability density function of the sEMG during isometric contraction in some cases is very close to the Gaussian distribution, it is frequently assumed to be Gaussian. As this assumption is not valid when the force is small, in this paper, we consider the non-Gaussian characteristics of the sEMG, and compute the second-, the third- and the fourth order statistics of the sEMG as its features. These features are used to classify four upper limb primitive motions, i.e., elbow flexion (EF), elbow extension (EE), forearm supination (FS), and forearm pronation (FP). We used the sequential forward selection (SFS) method to reduce the number of HOS features to a sufficient minimum while retaining their discriminatory information, and apply the Knearest neighbor method for classification. Our approach is robust against statistical variations in noise, and does not require additional computations compared to existing methods for providing high rates of correct classification of the sEMG, which makes it useful in devising real-time sEMG controlled prostheses.

One of the most important challenges in automatic speech recognition is in the case of difference between the training and testing data. To decrease this difference, the conventional methods try to enhance the speech or use the statistical model adaptation. Training the model in different situations is another example of these methods. The success rate in these methods compared to those of cognitive and recognition systems of human beings seems too much primary. In this paper, an inspiration from human being's recognition system helped us in developing and implementing a new connectionist lexical model. Integration of imputation and classification in a single NN for ASR with missing data was investigated. This can be considered as a variant of multi-task learning because we train the imputation and classification tasks in parallel fashion. Cascading of this model and the acoustic model corrects the sequence of the mined phonemes from the acoustic model to the desirable sequence. This approach was implemented on 400 isolated words of TFARSDAT Database (Actual telephone database). In the best case, the phoneme recognition correction increased in 16.9 percent. Incorporating prior knowledge (high level knowledge) in acoustic-phonetic information (lower level) can improve the recognition. By cascading the lexical model and the acoustic model, the feature parameters were corrected based on the inversion techniques in the neural networks. Speech enhancement by this method had a remarkable effect in the mismatch between the training and testing data. Efficiency of the lexical model and speech enhancement was observed by improving the phoneme's recognition correction in 18 percent compared to the acoustic model.

In this study 22 male Guinea Pigs, 4-6 months old, weighting 400-450 g were used. A computer controlled indentor system was used to apply a controlled pressure. The applied pressure was 291 mmHg for 3 hours over the trochanter region of animal hind limb. The animals were divided in three groups; in group 1, pressure was applied 3 hours continuously, in group 2, pressure was applied 90 minutes at two days and in group 3, Pressure was applied in two cycles of 90 minutes with 15 minutes rest between them. To study the biomechanical and histological changes, tissue was removed 7 days after pressure application. Uniaxial tensile test was performed at a deformation rate of 20 mm/min. In this test, the contralateral site on the experimental animal served as intra-animal control. Tissue biopsy was taken and stained with H&E and Trichorome for histological examination. Continuous pressure induced muscle necrosis. Also ultimate stress, stiffness, ultimate strain and area under the load-deformation curve decreased significantly. These results suggest that application of continuous pressure is the major cause of ischemia and necrosis of soft tissue.

In the first version of our Farsi Text-To-Speech (TTS) system, a Recurrent Neural Network (RNN) was used to generate prosody parameters (pitch contour, duration, energy and pause), and a Harmonic + Noise Model (HNM) speech synthesizer was used to concatenate the single units of diphones. To improve the performance of TTS, in this paper, two modifications are presented. In the first one is a neural-statistical hybrid model in which RNN plays the role of prosody parameterizer and the combination of decision trees and Gaussian Mixture Models (GMMs) gives the probability distributions of targets and transitions in each context a equivalent cluster. Another modification is about developing a unit selection speech synthesizer in which syllable is selected as the basic synthesis unit and, due to the first modification, an effective unit selection strategy is also conducted. To evaluate the performance of the system, the rating scales presented in the recommendation P.85 of the International Telecommunication Union (ITU) were used and the Mean Opinion Score (MOS) over six scales was achieved as 3.6.