Cochlear Implants Overview

Cochlear Implant

• A cochlear implant bypasses non-functioning hair cells in the ear and delivers weak electrical signals directly to the auditory nerve.
• Unlike hearing aids, which amplify sound, implants directly stimulate the auditory nerve.

Cochlea Sound Decoding

• High-frequency sounds (1500-20000 Hz) are detected by the base of the basilar membrane.
• Medium-frequency sounds (600-1500 Hz) are detected slightly above the base of the basilar membrane.
• Low-frequency sounds (200-600 Hz) are detected near the apex of the basilar membrane.

Cochlea Implant - Channels

• The cochlea implant has 18,800 pulses per second.
• White in the diagram represents a pulsed electrode.

Future of Cochlear Implants

• Bilateral cochlear implants provide directionality and improve speech recognition in noisy environments.
• Increasing the number of channels increases coverage for fine spectral information and improved speech and music appreciation.
• Reducing power allows for fully implantable devices.
• Performance is limited by surviving auditory nerve neurons; neuron regeneration is possible.

Bionic Eye

• A bio-electronic eye is an electronic device that replaces or enhances functionalities of the eye.

Regions of Implantation

• Retina
• Optic Nerve
• Lateral geniculate body
• Visual Cortex

Deep Brain Stimulation (DBS)

• DBS involves implanting electrodes within specific brain areas.
• These electrodes produce electrical impulses that regulate abnormal impulses within the brain.

DBS Applications

• Movement Disorders: Parkinson's Disease, Dystonia, Huntington's Disease, Essential Tremor
• Psychiatric Disorders: Obsessive-Compulsive Disorder (OCD), Severe Depression, Addiction (e.g., cocaine), Tourette Syndrome
• Epilepsy

Nerve Cuff

• A nerve cuff is a medical device that interacts with peripheral nerves.
• It is a tubular structure made of biocompatible materials (e.g., silicone).
• It encircles the nerve and contains electrodes strategically placed to deliver electrical impulses to the nerve.

Vagus Nerve Stimulation (VNS)

• A treatment for epilepsy using a stimulator connected to the left vagus nerve in the neck.
• Mild electrical stimulations help calm irregular electrical brain activity that causes seizures.
• Conditions treatable by implantable devices include epilepsy, depression, and stroke recovery.

Electromyography (EMG)

• EMG measures muscle response or electrical activity in response to nerve stimulation.
• It's used to detect neuromuscular abnormalities.
• Needles (electrodes) are inserted into muscles during the test.

Bionics

• Bionics applies natural methods and systems to the study and design of engineering systems.
• In medicine, bionics often refers to replacing body parts with mechanical or electrical enhancements.

Biofeedback

• Biofeedback involves acquiring real-time information about the body to self-regulate or change behaviours.

Neurofeedback Applications

• Future research is needed to fully understand the efficacy and effectiveness of neurofeedback in treating conditions such as depression, obsessive-compulsive disorder, traumatic brain injury, autism spectrum disorders, learning difficulties, migraines, and bipolar disorder.

Data Preprocessing

• Data preprocessing techniques used for brain signal data include:
  • Down sampling
  • Baseline Correction
  • Remove DC Offset
  • Filtering
  • Segmentation
  • Detrending
  • Feature extraction

Types of Time Domain Signals

• Static: Unchanging over a long period.
• Quasistatic: Nearly unchanging where changes are slow.
• Periodic Signal: Repeats itself regularly (e.g., sine wave, triangle wave).
• Repetitive Signal: Quasi-periodic, but not precisely periodic (e.g., ECG or arterial pressure wave).
• Transient Signal: A single, short-duration event.

Time vs. Frequency Relationship

• Infinitely continuous signals in the frequency domain are finite in the time domain.
• Conversely, signals infinitely continuous in the time domain are finite in the frequency domain.
• Mathematically, a signal cannot be both finite in time and frequency.

Spectrum and Bandwidth

• Spectrum: Range of frequencies in a signal.
• Absolute Bandwidth: Full width of the spectrum.
• Effective Bandwidth: Often just the bandwidth. Engineers utilize this to obtain the practical bandwidth of a signal.
• DC Component: Component with zero frequency.

Error with Digitalization

• Sampling Error: Sample rate needs to respect Nyquist theorem; sample rate at least 2 times the maximum frequency.
• Quantization Error: Digitizing involves signal levels based on bits in the Data Acquisition Card (DAC). Effects include "staircase" effect.

General Linear Model (GLM)

• GLM is an extension of multiple regression.
• Multiple regression only looks at one dependent variable.
• GLM allows analysis of multiple dependent variables in a linear combination.
• Techniques like ANOVA, t-tests, and F-tests are forms of GLM.

GLM and fMRI

• SPM (statistical parametric mapping) uses a mass univariate approach in fMRI.
• Each voxel is treated as a separate column vector, with the BOLD (blood oxygen level dependent) signal as Y at different time points in a given voxel.
• The design matrix (X) provides various component contributions, e.g., timing info (onsets and durations).
• An HRF (hemodynamic response function) describes the expected BOLD response over time. Other regressors may include realignment parameters.
• The parameters (β) are estimated to minimize the difference between observed (Y) and predicted data (Xβ). This is done to minimize the error (ε) using least squares techniques.

Biostatistical Concepts

• Sensitivity: Proportion of those with a condition who are correctly identified by the test.
• Specificity: Proportion of those without a condition who are correctly identified by the test.
• Positive Likelihood Ratio: Ratio of the probability of a positive test in those with the condition to the probability of a positive test in those without the condition.
• Negative Likelihood Ratio: Ratio of the probability of a negative test in those with the condition to the probability of a negative test in those without the condition.
• Positive Predictive Value: Probability of having the condition if the test is positive.
• Negative Predictive Value: Probability of not having the condition if the test is negative.
• Confidence Intervals: Range of values likely to contain the true value of a parameter.

Machine Learning

• Machine learning is a branch of Artificial Intelligence (AI), which is a branch of computer science.
• The idea in machine learning is to start with a generic model (e.g., a line), modify parameters until the minimum squared error is reached; this process is learning.
• An algorithm that finds the minimum squared error is called gradient descent.
• Once a model is learned, it can predict unseen values

Artificial Neural Networks

• Artificial neural networks (ANNs) consist of interconnected input/output units with weights.
• ANN learning is also known as connectionist learning.
• It is a case of supervised, inductive, or classification learning.
• Applications of ANNs include forecasting, manufacturing quality control, medicine (ECG data, RNA, DNA sequencing, drug development without animal testing), and process control/robotics.

Convolutional Neural Networks (CNNs)

• CNNs are feedforward neural networks in which weight multiplications are replaced by convolutions (filters).
• CNNs can directly process raw data (e.g., image, sound signals) without feature extraction.
• CNNs automatically learn features.

Deep Learning

• Deep learning allows for classification, unsupervised learning, denoising, generative models, games/puzzles, natural language processing, and scientific applications.

Activation Functions and Outputs

• Activation functions in neural networks are non-linear functions that determine the new activation level based on effective input and current activation.
• Types of activation functions include sigmoid, tanh, relu, leaky relu, maxout, elu.
  • Sigmoid: binary classification
  • Softmax: multi-class classification
  • Linear: general purpose regression

Deep Learning Successful Biomedical Applications

• Diabetic retinopathy detection
• Tumor detection (MRI, CT, X-ray)
• Skin lesion classification (clinical and dermoscopic images)
• Heart sound classification (normal vs. abnormal, murmur)
• Parkinson's disease (voice recording detection).

Support Vector Machines (SVMs)

• SVMs are an improvement upon perceptrons, addressing certain limitations.
• SVMs select a hyperplane that best separates different classes and maximizes the margin between classes.

Neuroinformatics

• Neuroinformatics integrates neuroscience, computer science, and information technology to advance our understanding of the brain.
• It involves the collection, storage, analysis, and visualization of neuroscience data.
• Neuroinformatics integrates neuromorphic engineering, computational neuroscience, and ontologies to quantify and visualize neuroscience datasets.

Fields Related to Neuroinformatics

• Computer Science (computational models of neuronal systems)
• Experimental Psychology (cognitive data, emotion, language)
• Medicine (aging, psychiatric diseases)
• Engineering (Brain-Computer Interfaces)
• Chemistry (atomic structure of nervous system)
• Mathematics (quantifying neuronal differentiation)
• Physical Sciences (physical processes within neural cells)
• Biology (chemical processes and molecular structure)

Challenges in Neuroinformatics

• Data Integration
• Data Privacy
• Standardization
• Computational Power
• Ethical Concerns

B2B Interface (B2BI)

• B2BI is an extension of neuroimaging technology, including BCIs and CBI.
• It allows data exchange between two brains through a BCI reading brain activity of one person and a CBI writing the data to the brain of another person.
• It has unidirectional and bidirectional collaboration types and applications range from rehabilitation and treatment to communication, collaboration, and synchronization..