A Deep Learning-Driven Multimodal Healthcare System for the Early Detection of Cervical Cancer

Screening parts of the human body for health analysis is a routine application of biomedical imaging. However, physicians do not rely only on imaging to arrive at a final diagnosis of the disease, but prefer clinical or laboratory results or reports to study the case along with the screened images. This study examined these dependent parameters to design an early prediction system for cervical cancer. A cervix-screened image of a patient is taken as input, along with clinical reports of the same patient. An image-based prediction is proposed by converting the original cervical image into the LAB color space before passing it through a two-channel Deep Convolution Neural Network (DCNN). The selection of the most suitable machine learning algorithm for accurate prediction based on clinical reports was ensured by focusing on recall. Logistic regression was the most effective technique in combining the two predictions for a final decision. The overall recall score of 95.83% shows the importance of the proposed method for early diagnosis.