Evolution and Tendency on the Feature Extraction Process for Diagnostic Aid in Healthcare

Introduction

Nowadays, in the current technological society, with the rise of tele-monitoring and continuous collection of user information, it is common to use wearable devices to monitor the physical, physiological or sleep activity of a user. These systems do not necessarily have to be medical devices or be provided by a professional, as there are multiple commercial and affordable solutions for all users such as activity measurement bracelets, smart watches, or even sensors integrated into the smartphones.

All these devices provide an easy way to obtain a large amount of data from multiple sensors with an improving accuracy year by year. All this collected data is also an inexhaustible source of information about the user's health state. Moreover, by carefully analyzing all this detailed information, very interesting conclusions can be drawn about the health conditions of patients, personalized plans of action and even the possibility of anticipating an anomalous event like falls or physical fatigue among others.

However, a medical professional cannot analyze this information in a personalized way due to the amount of data stored, which would require days or weeks to obtain objective conclusions from this data. That is why, in recent years, multiple commercial developments and research projects are being carried out based on diagnostic aid systems that can deal with this amount of information and providing objective results of the analysis performed. For this purpose, Big Data and Artificial Intelligence technologies are usually applied to this field.

In diagnostic support, programs based on Artificial Intelligence can analyze the high amount of data to assist healthcare professionals' decisions very quickly and with a high accuracy, allowing for further diagnostics as well as early diagnosis. In more detail, these intelligent models can use different types of input data, from medical images and physiological signals to demographic information and patient symptoms to detect and help in the diagnosis of neurological diseases as Alzheimer's disease (Asim et al., 2018), Parkinson disease (Gil & Johnson, 2009), or dementia (Amini et al., 2021); multiple types of cancer, like breast cancer (McKinney et al., 2020), lung cancer (Karhan & Tunç, 2016), liver cancer (Giordano et al., 2020) or skin cancer (Hekler et al., 2019), among others. AI-based diagnostic systems are also applied to detect chronic diseases, like diabetes (Jahangir et al., 2017), chronic obstructive pulmonary disease (Er & Temurtas, 2008), and chronic kidney disease (Qin et al., 2019); cardiovascular issues as strokes (Shoily et al., 2019), hypertension (Nour & Polat, 2020) or heart failures (Tripoliti et al., 2017); and also, infectious diseases as tuberculosis (Hrizi et al., 2022), pneumonia (Toğaçar et al., 2020) and, more recently, COVID-19 (Civit-Masot et al., 2020).