A Federated Learning Framework for Stenosis Detection

This study explores the use of Federated Learning (FL) for stenosis detection in coronary angiography images (CA). Two heterogeneous datasets from two institutions were considered: Dataset 1 includes 1219 images from 200 patients, which we acquired at the Ospedale Riuniti of Ancona (Italy); Dataset 2 includes 7492 sequential images from 90 patients from a previous study available in the literature. Stenosis detection was performed by using a Faster R-CNN model. In our FL framework, only the weights of the model backbone were shared among the two client institutions, using Federated Averaging (FedAvg) for weight aggregation. We assessed the performance of stenosis detection using Precision (P rec), Recall (Rec), and F1 score (F1). Our results showed that the FL framework does not substantially affects clients 2 performance, which already achieved good performance with local training; for client 1, instead, FL framework increases the performance with respect to local model of +3.76%, +17.21% and +10.80%, respectively, reaching P rec = 73.56, Rec = 67.01 and F1 = 70.13. With such results, we showed that FL may enable multicentric studies relevant to automatic stenosis detection in CA by addressing data heterogeneity from various institutions, while preserving patient privacy.

Performance assessment of medical and non-medical CPAP interfaces used during the COVID-19 pandemic

Background: At the beginning of 2020, a high number of COVID-19 cases affected Italy in a short period, causing a difficult supply of medical equipment. To deal with the problem, many healthcare operators readapted different masks to medical devices, but no experiment was conducted to evaluate their performance. The aims of our study were: to assess the performances of three masks and a CPAP helmet in their original configuration and after modifications, in the maintenance of mean pressure and half-amplitude variations using different PEEP valves and to analyse the impact of antibacterial (AB) or antibacterial-viral (ABV) pre-valve PEEP filters on the effective PEEP delivered to the patients. Four pressure ports were installed on each mask (three on helmet), mean values and half amplitudes of pressure were recorded. Tests were performed with any, AB, ABV filter before the PEEP valve. CPAP helmet was the most efficient interface producing more stable mean pressure and less prominent half-amplitude variations but the non-medical masks, especially after the modifications, maintained a stable mean pressure value with only a moderate increase of half-amplitude. The use of AB and ABV filters, produced respectively an increase of 2,23% and 16.5% in mean pressure, compared to no filter condition. CPAP helmet is the most reliable interface in terms of detected performance, but readapted masks can assure almost a similar performance. The use of ABV filters before the PEEP valve significantly increases the detected mean pressure while the AB filters have almost a neutral effect.