|
|
|||||
|
|
|||||
|
|
|
L2HAD™ (Lifelong Learning Heartbeat
Anomaly Detector) has been designed as an ECG trace analysis software with
the aim of indicating a level of abnormality in the heartbeat and thus
creating a priority list for the cardiologist. Therefore, the purpose of this
software is not to replace the cardiologist's evaluation, but only to define
an evaluation priority. L2HAD™ is based on ROCKET™ technology
and has been trained to recognize arrhythmia type anomalies (standard AAMI
EC57) and myocardial infarction. In reality these are the anomalies that are correctly
classified but the software is able to detect other types of anomalies
although these are classified in incorrect classes. The most generic output
of the software is a numerical value which corresponds to the criticality of
the heartbeat. The big difference between this type
of software and those based on DEEP-LEARNING technology is that this software
can continuously learn new ECG traces under the supervision of the
cardiologist. In fact L2HAD™ is based on a Neural Network model (contained in
the ROCKET™ technology) which has the property of continuous learning: this
functionality is not possible with a Deep-Learning type approach as it
requires a learning process that it must cycle multiple times over all the
previously learned data. L2HAD™ is based on a proprietary beat
extraction algorithm and uses the traditional Pan-Tompkins algorithm only in
special cases. This software is available as DLL for
Windows or as source code written in C language to be ported on any hardware
platform and any Operating System. L2HAD™ requires low computational and
memory capabilities. The ROCKET™ technology, on which
L2HAD™ is based, is functionally compatible with the Neuromem®
neuromorphic technology. It is therefore possible
to evaluate the porting of the library to be used with Neuromem®
technology for wearable applications. |
|
|
|
|
|
©2024_Luca_Marchese_All_Rights_Reserved Aerospace_&_Defence_Machine_Learning_Company VAT:_IT0267070992 NATO_CAGE_CODE:_AK845 Email:_luca.marchese@synaptics.org |
|
Contacts_and_Social_Media
|
