PREDICTION OF RHABDOMYOSARCOMA USING TRANSFER LEARNING ON PATHOLOGY AND RADIOLOGY IMAGES

Abstract: This paper presents the prediction of Rhabdomyosarcoma (RMS) using Transfer Learning model VGG19. While many machine learning techniques have been created to recognise the most common tumour types in histology images, considerably less is understood about the automatic categorization of tumour subtypes. The most prevalent soft tissue cancer in children, rhabdomyosarcoma (RMS), contains a number of subtypes, the most prevalent of which are embryonal, alveolar, and spindle cell. It is important to assign RMS to the appropriate subtype since different subtypes have been shown to react to various treatment modalities. Rhabdomyosarcoma is a common paediatric cancer of malicious soft-tissue tumour that affect 40% to 50% children more than adults. Due of the subtle differences in appearance of histopathology images, manual categorization needs a high level of knowledge and takes time. While many machine learning techniques have been created to recognise the most common tumour types in histology images, considerably less is understood about the automatic categorization of tumour subtypes. The most common sites of RMS tumor are head, neck, genitourinary tract, and extremities. The RMS prediction achieves 97.6% of accuracy for radiology images and 88.4% of pathology images byVGG19 model. Keyword: Transfer Learning, Machine Learning, Deep Learning, VGG 19