Abstract

Background and Aim: In the last decade, the battle against cancer has reached a new level: the merger between precision medicine and artificial intelligence (AI) models. AI is omnipresent in the healthcare system, providing fast, accurate diagnosis. By using microarrays of deoxyribonucleic acid and mass spectrometry, AI models such as artificial neural networks use their adaptive learning and nonlinear mapping properties to personalize their hyperparameters providing personalized diagnosis and treatment. Methods: Several novel adaptive neural networks are presented. Their diagnosis performance and computational speed is increased by having genomic knowledge embedded in their architecture. The fLogSLFN model uses the logistic regression to initialize the weights between the input and the hidden layer in a single hidden feedforward neural network (SLFN). This AI model can be applied when dealing a with a two-class decision problem. Because the fLogSLFN was limited to binary problems, we have extended it to the case of multiple classes by considering two approaches: a parallel LogSLFN and a cascaded LogSLFN. The parallel approach pLogSLFN uses the one-against-all method and thus transforms the multi-decision class problem into a two-class problem. The logSLFN is applied in parallel using the logistic regression to compute the weights for each class. The cascaded approach cLogSLFN uses the logistic regression to compute the weights for the first class, then eliminates them, and so on. Results: The fLogSLFN has proved to be successful in differentiating breast and lung cancer. The pLogSLFN and cLogSLFN have been applied on datasets concerning differentiating breast, kidney, colon, lung, prostate, and liver cancer. The statistical analysis revealed that the pLogSLFN outperforms the cLogSLFN. Conclusions: Diagnosing and treating cancer has changed through precision medicine. The DNA arrays contain crucial information and can be processed through AI models. Three novel AI models that provide fast and reliable diagnosis have been discussed.

Keywords

Artificial neural networks, Logistic regression, Cancer, Precision medicine