Advanced Non-Linear Generative Model with a Deep Classifier for Immunotherapy Outcome Prediction:
A Bladder Cancer Case Study


Francesco Rundo1,*, Giuseppe Luigi Banna2, Francesca Trenta3, Concetto Spampinato4,
Luc Bidaut5, Xujinog Ye5, Stefanos Kollias5, Sebastiano Battiato3


1 STMicroelectronics, ADG Central R&D Division, Catania, Italy
2 Medical Oncology Department, Queen Alexandra Hospital, Portsmouth, UK
3 IPLAB, University of Catania, Catania, Italy
4 PerCeiVe Lab, University of Catania, Catania, Italy
5 Computer Science Department, University of Lincolnshire, Lincolnshire, UK

*francesco.rundo@st.com, giuseppe.banna@nhs.net, francesca.trenta@unict.it, cspampinato@dieii.unict.it,
LBidaut@lincoln.ac.uk, XYe@lincoln.ac.uk, SKollias@lincoln.ac.uk, battiato@dmi.unict.it




ABSTRACT


Immunotherapy is one of the most interesting and promising cancer treatments. The effectiveness of immunotherapy drugs for the treatment of tumors has been confirmed by the comforting results in terms of long-term survival and significant reduction in toxicity compared to the classic chemotherapy approach. However, the percentage of patients eligible for immunotherapy is rather small and it is likely related to the limited knowledge of physiological mechanisms by which certain subjects respond to the treatment while others have no benefit. To address this issue, the authors propose an innovative approach based on the use of a cellular non-linear architecture with a deep downstream classifier to select and properly augment 2D features from chest-abdomen CT images of the patient toward providing prognostic information. The proposed pipeline has been designed to make it usable in an embedded system as innovative Point of Care. The authors report a case-study of the proposed solution applied to a specific type of aggressive tumor, namely, the Metastatic Urothelial Carcinoma (mUC). The performance evaluation (overall accuracy close to 93%) confirms the effectiveness of the proposed approach.