PuSH - Publikationsserver des Helmholtz Zentrums München

Sadafi, A. ; Moya Sans, L.M. ; Makhro, A.* ; Livshits, L.* ; Navab, N.* ; Bogdanova, A.* ; Albarqouni, S. ; Marr, C.

Fourier transform of percol gradients boosts CNN classification ofhereditary hemolytic anemias.

In: (2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI), 13-16 April 2021, Nice, France). 2021. 966-970
Preprint DOI

Hereditary hemolytic anemias are genetic disorders that af-fect the shape and density of red blood cells. Genetic testscurrently used to diagnose such anemias are expensive andunavailable in the majority of clinical labs. Here, we pro-pose a method for identifying hereditary hemolytic anemiasbased on a standard biochemistry method, called Percollgradient, obtained by centrifuging a patient’s blood. Our hy-brid approach consists on using spatial data-driven features,extracted with a convolutional neural network and spectralhandcrafted features obtained from fast Fourier transform.We compare late and early feature fusion with AlexNet andVGG16 architectures. AlexNet with late fusion of spectralfeatures performs better compared to other approaches. Weachieved an average F1-score of 88% on different classes sug-gesting the possibility of diagnosing of hereditary hemolyticanemias from Percoll gradients. Finally, we utilize Grad-CAM to explore the spatial features used for classification.

Weitere Metriken?
Zusatzinfos bearbeiten [➜Einloggen]
Publikationstyp Artikel: Konferenzbeitrag
Schlagwörter Image Classification, Deep Learning,Red Blood Cells, Percoll Density Gradients
Konferenztitel 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI)
Konferzenzdatum 13-16 April 2021
Konferenzort Nice, France
Quellenangaben Band: , Heft: , Seiten: 966-970 Artikelnummer: , Supplement: ,