Search

Background and objective: Automatic semantic segmentation of skin lesions is one of the most important medical requirements in the diagnosis and treatment of skin cancer, and scientists always try to achieve more accurate lesion segmentation systems. Developing an accurate model for lesion segmentation helps in timely diagnosis and appropriate treatment. Methods: In this study, a two-stage deep learning-based method is presented for accurate segmentation of skin lesions. At the first stage, detection stage, an approximate location of the lesion in a dermoscopy is estimated using deep Yolo v2 network. A sub-image is cropped from the input dermoscopy by considering a margin around the estimated lesion bounding box and then resized to a predetermined normal size. DeepLab convolutional neural network is used at the second stage, segmentation stage, to extract the exact lesion area from the normalized image. Results: A standard and well-known dataset of dermoscopic images, (ISBI) 2017 dataset, is used to evaluate the proposed method and compare it with the state-of-the-art methods. Our method achieved Jaccard value of 79.05%, which is 2.55% higher than the Jaccard of the winner of the ISIC 2017 challenge. Conclusion: Experiments demonstrated that the proposed two-stage CNN-based lesion segmentation method outperformed other state-of-the-art methods on the well-known ISIB2017 dataset. High accuracy in detection stage is of most important. Using the detection stage based on Yolov2 before segmentation stage, DeepLab3+ structure with appropriate backbone network, data augmentation, and additional modes of input images are the main reasons of the significant improvement.

Background & Objective: Major management decisions in organizations not only in the present but also in the future have a profound impact on different aspects of the organization. A slight mistake in making decisions may lead to the loss of resources of the organization, including financial and human resources. In the present study, we evaluated the problem of choosing the most convenient location for the construction of hospitals and health centers as one of the most important issues in the field of health. Regarding the numerous factors in decision making and the myriad of possible solutions to this problem and also disability of human in solving such problems, a genetic optimization algorithm has been used to calculate the best location for the construction of hospitals. Methods: This study was simulated according to the actual conditions which may exist in a city. Given the existence of a city with N × N dimensions and having several hospitals and health centers in the city, the issue was raised for the construction of three hospitals. Important factors which could influence the decision making were health status, referring times and land prices. Furthermore, the most proper locations for the construction of three hospitals were calculated using the genetic algorithm. Results: Three characteristics including the level of health, referring times and land prices were randomly assigned to all urban areas. The coordinates of available health centers in the city were also identified. Another point was the lack of proximity of hospitals in the city. Setting the threshold of 0.2 units for the minimum distance between hospitals (current and new), this restriction was applied. After performing the algorithm with the governing conditions, three optimal points were found. Conclusion: Considering the importance of locations for the construction of hospitals and health centers in the city and the existence of various factors for selecting the most appropriate place, application of strategies and algorithms which may be helpful in finding the best solution among the myriad of solutions in inevitable. According to the fact that human beings alone or by simple mathematical methods are not capable of taking all the features together and examine the search space to find the best result, we achieved the best solution in the city by setting the parameters of the genetic algorithm and taking into account all important factors.

Background and objectives: Investigatingg the mortality in a population has been considered as one of the appropriate methods of health detection. Although, there are some problems such as lack of confidence in accuracy measurement and quality of data collection. Establishment of death registration systems and using international classification codes of diseases, and also mortality data integrating by responsible organizations have solved great parts of the previous problems. In this study, considering a set of parameters, the study population was divided into two groups: deceased under one year (infants) and over one year (adults). Then both groups were clustered using the K-means method to identify different groups. Hidden models and useful patterns were also discovered using decision tree algorithms. Finally, a neural network algorithm was used to show the ranking of attributes in order of their importance. Methods: In this research, data of 12,865 deceased individuals in Golestan province since 2007 to 2009 is studied. The data has been obtained from the Health Center of Golestan province. The main characteristics used in this study are: deceased age, gender, cause of death, place of residence and place of death. K-means algorithm is used to cluster data. The decision tree algorithms and neural networks algorithm were also used for classification. Finally, results and rules were extracted. Due to different natures of causes of death in infants and adults, studying on these different groups is performed separately. Results: In clustering phase, the optimal number of clusters is obtained by Dunn index; eight clusters for infants and seven clusters for adults were obtained. Among four decision-tree algorithms (C5.0, QUEST, CHAID and CART), C5.0 algorithm with high correction rate, 77.37% in infants data and 96.86% in adults data was the best classifier algorithm. Age, gender and place of death were the most important variables that were detected by neural network algorithm. Conclusion: In the present study, the collected mortality data was clustered by considering the effective factors and the standard of International Classification of Diseases. The hidden patterns of mortality for infants and adults were extracted. Due to the explicit nature and the intelligibility of the decision tree algorithms, the results and extracted rules are very useful for specialists in this field.

Background & Objective: Prediction of health status in newborns and also identification of its affecting factors is of the utmost importance. There are different ways of prediction. In this study, effective models and patterns have been studied using decision tree algorithm. Method: This study was conducted on 1,668 childbirths in three hospitals of Shohada, Omidi and Mehr in city of Behshahr. Variables such as baby's gender, birth weight, birth order, maternal age, maternal history of illness, gestational diseases, type of delivery, reason of caesarean section, maternal age, family relationship of father and mother, mother's blood type, mother's occupation and blood pressure and place of residence were chosen as predictive factors of decision tree categorization method. The health status of the baby was used as a dependent dual-mode variable. All variables were used in clustering and correlation rules. Prediction was done and then compared using 4 decision-tree algorithms. Results: In the clustering method, the optimal number of clusters was determined as 8, using the Dunn index measurement. Among all the implemented algorithms of CART, QUEST, CHAID and C5.0, C5.0 algorithm with detection rate of 94.44% was identified as the best algorithm. By implementing the Apriori algorithm, strong correlation rules were extracted with regard to the threshold for Support and Confidence. Among the characteristics, maternal age, birth weight and reason of caesarean section with the highest impacts were found as the most important factors in the prediction. Conclusion: Due to the simple interpretation of the decision tree and understandability of the extracted rules derived from it, this model can be used for (most individuals) professionals and pregnant women at different levels.