l. peng

Traditional diagnostic methods are limited in accuracy when detecting maxillary sinus fungal balls, leading to a higher risk of misdiagnosis or missed diagnosis. This study focuses on a deep learning-based algorithm for assisting in the localization and diagnosis of maxillary sinus fungal balls, addressing the limitations of conventional diagnostic procedures.

Axial CT imaging data of maxillary sinus were collected from 107 patients, including 47 cases of maxillary sinus fungal balls, 30 cases of other maxillary sinus lesions and 30 cases of healthy maxillary sinus, based on which, a dataset was constructed and a two-stage assisted diagnosis algorithm consisting of a classification and detection model was established. In the first stage, slices containing maxillary sinus were classified and selected. In the second stage, the selected slices were detected to diagnose and localize the fungal ball lesions in the maxillary sinus.

The accuracy of the classification model was 92.71%, the mAP and AP50 of the detection model were 0.73 and 0.76, respectively, and the accuracy of the algorithm for the diagnosis of maxillary sinus fungal balls was 84.4%.

It is feasible to develop a two-stage auxiliary diagnosis method for maxillary sinus fungal ball based on deep learning.

Online handwritten Urdu character recognition is one of the key technologies for intelligent interface on smart phones and touch screens. It is a challenging research topic as Urdu script has many similar character groups. A novel similar character discrimination method for online handwritten Urdu character recognition is proposed in this paper which includes pre-classification, feature extraction and fine classification process. The pre-classifier enables the discrimination of similar characters by putting them in distinct smaller subsets according to stroke number and diacritics. Then structural features and wavelet features are extracted. Finally, Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Recurrent Neural Network (RNN) classifiers are compared for fine classification within subsets. Results of RNN classifier without using the proposed pre-classifier and features have also been obtained to check the end-to-end capability of the RNN classifier. Experimental results show that the proposed method is efficient and achieves an overall accuracy of 96% on a large-scale self-collected dataset. It is feasible to extend this method for other Arabic scripts.

Nowadays, some burgeoning issues occurring in green and sustainable supply chain face an increasing number of tackles,  including uncertain factors and unobserved data, which makes it even more complicated to assess supply chain efficiency. To address this issue, this paper applies uncertainty theory to two-stage network DEA  intending to deal with inaccurate data. Moreover, the bi-echelon supply chain generally suffers asymmetric power among involved members. Therefore, this paper proposes attrition rate (D) and fulfillment rate (G) in the first and second stages respectively so as to reveal the  bi-echelon supply chain within different leader-follower patterns. The first two models assume the upstream firm is the leader while the last two models regard the downstream firm as the leader. We find out that the results of the evaluation vary under divergent patterns accordingly by running a numerical example. Adopting proposed models can help decision-makers of the multistage supply chain make decisions more effectively and avoid possible mistakes.