ISSN 1866-8836
Клеточная терапия и трансплантация

Artificial neural network in total survival predicting of multiple myeloma patients

Maria V. Markovtseva1, Vadim V. Shishkin2

1 Ulyanovsk State University, Ulyanovsk, Russia
2 Ulyanovsk State Technical University, Ulyanovsk, Russia


Contact: Maria V. Markovtseva
E-mail: mmark7@yandex.ru

Download PDF version

Cellular Therapy and Transplantation (CTT)
Volume 8, number 3
Contents 

Summary

At the present stage, the prediction of total survival (TS) in multiple myeloma (MM) is usually carried out by the ISS staging system (2005). In real clinical practice, the parameters can significantly differ from the expected, while some patients overcome it, and some do not reach it.The more accurate prediction of the patients TS will optimize the therapeutic tactics choice and take into account the patient`s individual characteristics. The latter reflects the personalized medicine principles, which are the basis of modern trends in therapeutic science. The aim of this work was a study of artificial neural network (ANN), in order to predict TS in patients with MM, because the ANN has the properties of a universal classifier, clusterization and can be used for regression analysis.

Methods

There were examined 135 patients MM I-III stage with known TS data. At the time of diagnosis, gender, age, Charlson comorbidity index were taken into account, and biochemical parameters such as total protein, albumin, ß2-microglobulin, creatinine, glomerular filtration rate by MDRD, urea, uric acid, lactate dehydrogenase, alanine aminotransferase, aspartate aminotransferase, total bilirubin, indirect bilirubin, glucose were studied. It was necessary to develop the ANN for TS predicting on available clinical data. The ANN based on simple perceptrons was chosen for the study, implemented in the language of artificial intelligence Python. As input, the two Excel spreadsheets were used, storing the initial data on the clinical performance of patients and data on the clinical performance of patients with known TS.

As an output document for the information system, an Excel spreadsheet was also used, in which, as a result of the ANN work, the prognostic value of the patient’s TS was determined. The information system is implemented in the form of a doctor automated place, with the ability to transfer information to the hospital information system.

Results

The two modes of operation were implemented in the system: training and forecasting. In the training mode, the results of clinical data and TS were fed to the input of the neural network and neuron weights were adjusted. ANN training was conducted on all known patients (135 people) and was repeated 100 thousand times to more accurately adjust the significance of clinical parameters affecting TS. In the forecast mode, clinical data results were fed to the input of the neural network and forecasts were formed. The ANN experimental studies are showing the results with artificial neural network (Table 1).

Table 1. The artificial neural network for myeloma: experimental testing

Markovtseva_tab01.jpg

Conclusions

The experiment showed more accurate TS prediction using ANN, compared to the currently adopted ISS system. In addition, the ANN provides for the study of existing relationships on ready-made models, does not require assumptions of the main distribution of the population, and is able to work with incomplete and fuzzy data. The use of intellectual information technologies opens up new opportunities in the study of dynamic problems in the field of medicine.

Keywords

Artificial neural network, multiple myeloma, total survival, prognosis.


Back to the list