autoSCORE (V 2.0.0)

autoSCORE V2.0 is a software-only decision support tool that analyzes EEG recordings to identify and categorize abnormalities into four predefined types: focal epileptiform, generalized epileptiform, focal non-epileptiform, and diffuse non-epileptiform. It assists qualified medical professionals by marking sections of previously acquired EEG data that may contain abnormalities and providing probability assessments, requiring independent clinical review before any diagnostic conclusion.

autoSCORE V2.0 uses a locked convolutional neural network algorithm trained with deep learning principles, unlike the primary predicate device (encevis) which does not assess non-epileptiform abnormalities or provide probability categorization. The device integrates with compatible EEG reviewing software through an integration layer, receives raw EEG data and metadata as input, and returns annotated results without storing data. It does not provide diagnostic conclusions, seizure detection, burst suppression analysis, or quantitative measures.

Not stated in this summary. The document references FDA Guidance for Industry on Software Contained in Medical Devices and mentions bootstrap resampling for statistical analysis, but does not cite specific ISO, IEC, or ASTM standards.

Clinical validation demonstrated autoSCORE V2.0 achieves similar or superior positive predictive value (PPV) compared to predicate devices: recording-level abnormal detection PPV of 0.969 versus encevis 0.920, and marker-level PPV of 0.82 for high-probability markers (90-100% range) versus encevis 0.257. The additional detection of non-epileptiform abnormalities does not introduce additional safety risks because results are independently reviewed by qualified clinicians, and performance metrics show comparable accuracy and specificity to predicates across multiple abnormality types and patient subgroups.