Muzna Nanaa

Dr Muzna Nanaa is a Research Associate and Honorary Clinical Fellow at the University of Cambridge, Department of Radiology. Under the supervision of Professor Fiona Gilbert, she has led many research projects over the past three years, and participated in UK-wide and international trials, including BRAID (Breast Screening – Risk Adaptive Imaging for Density) and MyPEBS (My Personalised Breast Screening). Dr Nanaa’s research interests include the use of artificial intelligence for breast cancer detection, quantification methods for contrast-enhanced mammography (CEM), and estimation of breast tumour volume doubling time. Additionally, she has worked on mechanistic studies investigating the tumour microenvironment in breast cancer and its role in chemotherapy response, using multi-parametric MRI and positron emission tomography (PET).

In 2023, Dr Nanaa in collaboration with Dr Iris Allajbeu (Senior Clinical Research Associate and Honorary Consultant, Department of Radiology, University of Cambridge) received a grant from the British Society of Breast Radiology (BSBR) to investigate the use of radiomics on CEM images. The project aims to leverage advanced image analysis techniques to extract quantitative features from CEM images in order to enhance breast lesion characterisation. This approach is expected to improve the accuracy of CEM interpretation and assist in clinical decision-making.

Classifying breast lesions using radiomic phenotypes from contrast-enhanced mammography - 2023 BSBR Bursary Award

Radiomic signatures derived from contrast-enhanced mammography (CEM) present a promising approach for the characterisation of breast lesions. Radiomics allows the extraction of numerous quantitative features from CEM images, such as texture, shape, and intensity, that are imperceptible to the human eye. Previous studies have shown that malignant lesions exhibit distinct radiomic patterns compared to benign ones. For instance, malignant tumours often display higher heterogeneity and specific textural features indicative of aggressive growth, while benign lesions typically present with more uniform characteristics.

The aim of this study was to develop radiomic signatures (or phenotypes) from CEM examinations to differentiate between benign and malignant breast lesions. We analysed 143 lesions classified as BI-RADS 2-5, extracting 102 radiomics features from both low-energy and recombined images, focusing on shape, first-order and second-order (texture) families. After correcting features for lesion size dependency and selecting features via correlation analysis, 11 radiomic features were identified (shape, n=3; first-order, n=2, second-order, n=6).

Using unsupervised hierarchical clustering, we identified ten clusters (signatures) which were significantly associated with lesion histology. Notably, malignant lesions were linked to higher values of certain textural features (Id, HighGrayLevelEmphasis, LongRunEmphasis) compared to benign lesions. In contrast, benign lesions showed less variation in these textures, with a subset of lesions being linked with first-order features like 10Percentile and InterquartileRange.

Future research will focus on correlating the derived radiomic signatures with specific histological characteristics of lesions, such as background parenchymal enhancement, tumour histology, grade, or molecular subtype, and validating these signatures in a larger cohort of lesions.

Speakers

Muzna Nanaa