Student advancements at the TenD Innovations Centre
16 June 2021
The University of Buckingham’s School of Computing and TenD Innovations in Shanghai established a collaborative partnership in 2018. Since then research students at the TenD Centre have been developing machine and deep learning solutions for cancer detection. As well as recognition from ultrasound images and other computer vision applications. Our dedicated students have had many papers accepted at national and international conferences.
Jehan Ghafuri presented her paper: “Sensitivity and Stability of Pretrained CNN Filters” at SPIE, the international society for optics and photonics conference in April 2021. The paper investigates the effect of filters on convolutional neural network (CNN) models. Determining the stability of decision making by the deep learning models. Resulting in more stable and reliable deep learning diagnostic models for lesion recognition. Ghafuri’s investigation used a topological approach to identify relevant filter properties.
Tahir Hassan’s paper: “Towards Optimal Cropping: Breast and Liver Tumour Classification Using Ultrasound Images” was also presented at SPIE 2021. It investigates the effect of background and surrounding tissue on the accuracy of lesion recognition. Using ultrasound images of breast and liver tissue. This tailored approach took relevant ultrasound images to analyse lesions and tumours.
Mohammed Ahmed’s paper: “Improving Generalization of ENAS-based CNN Models for Breast Lesion Classification from Ultrasound Images” has been accepted for the 25th UK Conference on Medical Image Understanding and Analysis (MIUA). To be held at the University of Oxford in July 2021. Ahmed’s paper investigates “model overfitting” scenarios. With automatically optimized CNN models for breast lesion classification, using ultrasound images. The research aims to develop more robust and reliable CNN models. For use in different medical centres, using ultrasound images from various machines. This work extends Ahmed’s paper: “An ENAS Based Approach for Constructing Deep Learning Models for Breast Cancer Recognition from Ultrasound Images”. Presented at the Medical Imaging with Deep Learning (MIDL) conference 2020.
Ali Eskandari’s paper: “Towards Linking CNN Decisions with Cancer Signs for Breast Lesion Classification from Ultrasound Images”, has been accepted for the MIUA 2021 conference. Eskandari’s paper investigates the current uses of CNN in breast cancer diagnosis. Aiming to understand how models make decisions, before using them as a clinical tool. This research is the first which directly identifies the link between the CNN model’s decision and calcification malignancy a sign of breast lesion classification.
Fakher Mohammad’s paper: “A Generic Approach for Automatic Crack Recognition in Building’s Glass Facade and Concrete Structures”, presented at the 13th International Conference on Digital Image Processing, held in Singapore, May 2021. It addresses an important and serious issue of high buildings in modern cities. Where cracks in glass and concrete can lead to serious and potentially fatal incidents. The paper proposes using machine learning solutions installed on drones for real-time detection. Allowing timely intervention, preventing serious incidents from happening. This work extends Mohammad’s paper: “Automatic glass crack recognition solution for inspection of high building façade”. Which was presented at the SPIE 2020 conference.
Students in the School of Computing have many opportunities to further their research. Including, submitting papers to national and international conferences.
Check out upcoming School of Computing events and seminars.