Publications

Artificial Intelligence enabled automated reception to perform as a human receptionist to avoid face-to-face interaction among mass people regarding their daily service in the current pandemic. Inspired by their current problem of mass congestion, our proposed AI-based Smart Reception can authenticate users and interact in Bangla language with humans by responding to university-domain-related queries, resulting in better business service and outcomes. We used OpenFace face recognition for authentication, having an accuracy of 92.92% with 1×10−5 second training time for a new image by saving the image dataset as a collection of an array file. The Interaction Recognition system consists of three modules: Automatic Speech Recognition (ASR), Interactive Agent, and Text-to-Speech (TTS) Synthesis. We used the OpenSLR- Large Bengali ASR Training Data to train the Deep Speech 2 model for ASR with a Word Error Rate (WER) of 42.15%. We tested our developed database management architecture for the Interaction Recognition system with the three-step evaluation using BERT sentence transformer (paraphrase-mpnet-base-v2) that provided satisfactory responses with 92% accuracy, increasing the receptionist performance significantly. TTS module relays on WavNet gTTS model. Our research also demonstrated that a developed AI-based system could be an adaptive solution for any domain-specific reception system responsible for systematic and efficient customer service offline and online.

The research study presents an architecture of HumanRobot Interaction (HRI) based Artificial Conversational Entity integrated with speaker recognition ability to avail modern healthcare services. Due to the Covid-19 pandemic, the situation has become troublesome for health workers and patients to visit hospitals because of the high risk of virus dissemination. To minimize the mass congestion, our developed architecture would be an appropriate, cost-effective solution that automates the reception system by enabling AI-based HRI and providing fast and advanced healthcare services in the context of Bangladesh. The architecture consists of two significant subsections: Speaker Recognition and Artificial Conversational Entities having Automatic Speech Recognition in Bengali, Interactive Agent, and Text-to-Speech-synthesis. We used MFCC features as the linguistic parameters and the GMM statistical model to adapt each speaker’s voice and estimation and maximization algorithm to identify the speaker’s identity. The developed speaker recognition module performed significantly with 94.38% average accuracy in noisy environments and 96.27% average accuracy in studio quality environments and achieved a word error rate (WER) of 42.15% from RNN based Deep Speech 2 model for Bangla Automatic Speech Recognition (ASR). Besides, Artificial Conversational Entity performs with an average accuracy of 98.58% in a small-scale real-time environmen.

Osteosarcoma is an osseous tumor that occurs in the metaphyseal area around the knee accounts for roughly 20% of bone cancers mostly affects patients younger than 20 years. Early diagnosis of osteosarcoma cancer can pave the way for an unlimited choice of therapy opportunities. Moreover, pathological estimation of necrosis and tumor cells determines the future intensity of chemotherapy radiation to apply to patient. The biopsy confirms the diagnosis and divulges the grade of the tumor, necrotic, and non-tumor cells. Due to a lack of radiologists in third world countries like Bangladesh, it is extremely difficult to diagnose cancer in the early stage. Moreover, to identify the chemotherapy effect during the chemotherapy period, multiple radiologists are required which is quite expensive for most cancer hospitals. In this paper, a Sequential Recurrent Convolutional Neural Network (RCNN) model consisting of CNN and bidirectional Gated Recurrent Units (GRU) is proposed, which performs exceptionally well with small numbers of histopathological osteosarcoma Haematoxylin and Eosin (H & E) stained images despite having the over-fitting problem, heterogeneity, intra-class variation, inter-class similarity, crowded context, the irregular shape of the nucleus and noisy data. Performance of the is compared with that of AlexNet, ResNet50, VGG16, LeNet and SVM models with the histopathological image dataset on osteosarcoma.

The research study presents an architecture of HumanRobot Interaction (HRI) based Artificial Conversational Entity integrated with speaker recognition ability to avail modern healthcare services. Due to the Covid-19 pandemic, the situation has become troublesome for health workers and patients to visit hospitals because of the high risk of virus dissemination. To minimize the mass congestion, our developed architecture would be an appropriate, cost-effective solution that automates the reception system by enabling AI-based HRI and providing fast and advanced healthcare services in the context of Bangladesh. The architecture consists of two significant subsections: Speaker Recognition and Artificial Conversational Entities having Automatic Speech Recognition in Bengali, Interactive Agent, and Text-to-Speech-synthesis. We used MFCC features as the linguistic parameters and the GMM statistical model to adapt each speaker’s voice and estimation and maximization algorithm to identify the speaker’s identity. The developed speaker recognition module performed significantly with 94.38% average accuracy in noisy environments and 96.27% average accuracy in studio quality environments and achieved a word error rate (WER) of 42.15% from RNN based Deep Speech 2 model for Bangla Automatic Speech Recognition (ASR). Besides, Artificial Conversational Entity performs with an average accuracy of 98.58% in a small-scale real-time environmen.

Most of the patients today who face health problems, initially take advice from unprofessional or people with no knowledge that makes them more vulnerable. In many occasions, doctors also get confused with identifying actual disease. This might happen as they usually identify disease based on their limited experience. Moreover, general patient selects doctor according to their will and with no knowledge about the disease that may need specialist doctor. But some disease cannot be confirmed without a specialized doctor. Therefore, this paper proposes a Machine Learning based disease symptom analysis technique for assisting the patients seeking proper treatment by selecting accurate medical department using the symptom that they can easily recognize. Proposed framework will use machine learning technique to select a medical department based on the joint consideration of various disease symptoms of the patient. We investigate our proposed framework by using 9 different supervised machine learning techniques. Performance of framework for identifying appropriate medical department under the machine learning techniques is thoroughly investigated and compared. This framework can be used for telemedicine platform or in automated hospital management sector. This may create a path of enormous development in health care sector.