The purpose of the present work is to analyze various models present in scientific literature to calculate the priority index for the replacement of electromedical devices in public hospitals or private clinics. The replacement priority is determined following the evaluation of several parameters that indicate technological or physical obsolescence of a given device. After analyzing various papers, three models based on different parameters and employing different algorithms and calculation methods were selected and applied in the same context: the equipment fleet of the ULSS3 hospital organization in Veneto. For each device, the output of the three models is calculated using the parameters proposed in the original papers and the corresponding values are computed. Thresholds are then applied to the obtained IPS values to classify each device as “Obsolete”, “Requiring Further Investigation”, or “Non-Obsolete”. The differences in the outputs of the models are due to the use of different parameters and thresholds, which make a model more or less selective. An algorithm is then created to combine the three outputs, obtaining a classification of the equipment fleet into four different urgency levels: “High”, “Middle”, “Low” and “None”. This classification is the basis for the development of an artificial neural network, to explore a potential application of neural networks in the context of prioritizing the replacement of electromedical devices. The neural network receives all the parameters used by the different models in input, without weights or coefficients, and predicts the urgency class in which the device was classified by the algorithm. The use of a neural network introduces variability into the results, but since the net is supported by a good accuracy and a small number of justified errors, it ensures a reduced complexity for possible future applications to other datasets. The neural network results are then evaluated through the classification metrics and the use of some Python libraries like SHAP and Alibi in order to understand the parameters that influenced the final choice of the assigned class for each medical device.
Il progetto si pone come scopo l’analisi di diversi modelli presenti in letteratura per valutare l’obsolescenza dei dispositivi medici e la loro priorità di sostituzione calcolata attraverso dei parametri indicativi della condizione dell’apparecchio. In seguito all’analisi di diversi articoli scientifici, vengono scelti tre modelli basati su un numero variabile di parametri e con algoritmi e metodi di calcolo differenti ed applicati al medesimo contesto, ovvero i dispositivi elettromedicali in dotazione all’Azienda Ospedaliera ULSS3 del Veneto. Per ogni apparecchio presente in azienda viene calcolato l’output dei tre modelli utilizzando i parametri proposti. In seguito, vengono applicate delle soglie ai valori ottenuti di IPS (indice di priorità di sostituzione) per determinare se ciascun apparecchio viene classificato come “Obsoleto”, “Da approfondire” o “Non Obsoleto”. I differenti risultati nella classificazione basata sui tre modelli sono attribuibili ai diversi parametri utilizzati e alle diverse soglie applicate per la classificazione, che rendono un modello più o meno selettivo. Viene successivamente creato un algoritmo che permette di unire il risultato dato dai tre modelli per ciascun apparecchio, ottenendo una classificazione in quattro fasce d’urgenza: “Massima”, “Media”, “Minima” e “Nessuna”. La classificazione ottenuta grazie a questo algoritmo viene utilizzata come base per l’addestramento di una rete neurale, che consente di valutare una possibile applicazione delle reti al contesto della priorità di sostituzione dei dispositivi elettromedicali. La rete neurale riceve come input tutti i parametri utilizzati dai diversi modelli, inseriti senza pesi o coefficienti, e predice la classe di urgenza in cui è stato classificato il dispositivo in seguito all’utilizzo dell’algoritmo. L’utilizzo di una rete neurale introduce della variabilità ma, poiché supportata da una buona accuratezza e pochi errori giustificati, permette di semplificare il processo per eventuali utilizzi futuri e diverse applicazioni. I risultati ottenuti dalla rete vengono esaminati attraverso una valutazione delle metriche ottenute e con l’ausilio di librerie di Python quali SHAP e Alibi.
Sviluppo di un modello basato su reti neurali per la valutazione dell'obsolescenza e sostituzione prioritaria di dispositivi elettromedicali in abito ospedaliero
VESCOVO, GIULIO
2023/2024
Abstract
The purpose of the present work is to analyze various models present in scientific literature to calculate the priority index for the replacement of electromedical devices in public hospitals or private clinics. The replacement priority is determined following the evaluation of several parameters that indicate technological or physical obsolescence of a given device. After analyzing various papers, three models based on different parameters and employing different algorithms and calculation methods were selected and applied in the same context: the equipment fleet of the ULSS3 hospital organization in Veneto. For each device, the output of the three models is calculated using the parameters proposed in the original papers and the corresponding values are computed. Thresholds are then applied to the obtained IPS values to classify each device as “Obsolete”, “Requiring Further Investigation”, or “Non-Obsolete”. The differences in the outputs of the models are due to the use of different parameters and thresholds, which make a model more or less selective. An algorithm is then created to combine the three outputs, obtaining a classification of the equipment fleet into four different urgency levels: “High”, “Middle”, “Low” and “None”. This classification is the basis for the development of an artificial neural network, to explore a potential application of neural networks in the context of prioritizing the replacement of electromedical devices. The neural network receives all the parameters used by the different models in input, without weights or coefficients, and predicts the urgency class in which the device was classified by the algorithm. The use of a neural network introduces variability into the results, but since the net is supported by a good accuracy and a small number of justified errors, it ensures a reduced complexity for possible future applications to other datasets. The neural network results are then evaluated through the classification metrics and the use of some Python libraries like SHAP and Alibi in order to understand the parameters that influenced the final choice of the assigned class for each medical device.File | Dimensione | Formato | |
---|---|---|---|
Vescovo_Giulio.pdf
accesso riservato
Dimensione
3.77 MB
Formato
Adobe PDF
|
3.77 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/78078