Pregled bibliografske jedinice broj: 953341
SEPSIS PREDICTION USING ARTIFICIAL INTELLIGENCE ALGORITHMS
SEPSIS PREDICTION USING ARTIFICIAL INTELLIGENCE ALGORITHMS // Internarional Conference on Inovative Technologies (IN-TECH 2018) / Car, Zlatan ; Kudláček, Jan (ur.).
Zagreb: Tehnički fakultet Sveučilišta u Rijeci, 2018. str. 47-50 (predavanje, međunarodna recenzija, cjeloviti rad (in extenso), znanstveni)
CROSBI ID: 953341 Za ispravke kontaktirajte CROSBI podršku putem web obrasca
Naslov
SEPSIS PREDICTION USING ARTIFICIAL INTELLIGENCE ALGORITHMS
Autori
Brnić, Mateo ; Čondrić, Emanuel ; Blažević, Sebastijan ; Anđelić, Nikola ; Borović, Emanuel ; Car, Zlatan
Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni
Skup
Internarional Conference on Inovative Technologies (IN-TECH 2018)
Mjesto i datum
Zagreb, Hrvatska, 05.09.2018. - 07.09.2018
Vrsta sudjelovanja
Predavanje
Vrsta recenzije
Međunarodna recenzija
Ključne riječi
artificial neural networks, machine learning algorithms, sepsis, support vector machines
Sažetak
Sepsis is a serious, life-threatening condition. It is an extreme reaction of the body to an infection.When the body starts fighting that infection, it damages its own organs and tissue. Septic shock is the last stage of sepsis, and it leads to dangerously low blood pressure and sometimes even death. Patients with sepsis are usually treated in intensive care units (lCUs). In cooperation with the Clinical Hospital Center in Rijeka (Croatia), and its ICU, data was collected for use in different machine learning algorithms, so that predictions of sepsis occurrence in patients could be recognized by those algorithms. The data consists of 24 different parameters for each patient, where the 24th is a simple binary value representing sepsis occurrence for a patient. An artificial neural network (ANN) is being trained on this data, with 23 of the medical parameters used as inputs and binary sepsis occurrence as outputs. ANNs testing results show that connections between the input parameters and sepsis occurrence clearly exist, and with each new batch of data from the ICU, the accuracy of the neural network increases and therefore the predictions get better. Also, two other machine learning algorithms (classification methods) are used on the same data, so that results could be compared between the three. Support Vector Machine (SVM), based on the aforementioned data, builds a model that classifies patients to one of the two categories - patients with sepsis or patients without sepsis. Unlike SVM, K-Nearest Neighbors (KNN) method uses only part of the training data and accordingly classifies new patients.
Izvorni jezik
Engleski
Znanstvena područja
Računarstvo, Kliničke medicinske znanosti
