A Salty Kiss initiated Chip: Encoding an unbiased Disease Diagnosis in a Neuromorphic platform — Supervised Binary Method to provide point-of-care services to Cystic Fibrosis Patients
The field of Artificial Intelligence is rapidly advancing, offering valuable support to clinicians in analyzing patient symptoms, classifying disease subtypes, and predicting drug effectiveness. For instance, supervised learning can forecast the progression of Montreal Cognitive Assessment scores in Parkinson’s disease patients, while innovative convolutional neural networks integrated into CT and MRI imaging processes intelligently handle classification and segmentation tasks. However, challenges such as limited on-call appointments in rural areas and expensive consultant fees can place undue pressure on patients’ families. Consequently, having an affordable and intelligent chip for monitoring disease progression and treatment outcomes at one’s disposal becomes highly convenient. Daily point-of-care systems, ranging from paper-based platforms to microfluidic chips, provide practical solutions, exemplified by common nasal screening tests during the peak of the COVID-19 season. Importantly, the non-invasiveness of these devices extends their utility to newborns, facilitating rapid sweat tests. This news article provides an interesting medical case, namely cystic fibrosis, and evaluates a chip-enabled point-of-care system doped with a hardware neural network.
What is Cystic Fibrosis?
Cystic Fibrosis, an autosomal recessive disease with fatal consequences, was first identified through a touching observation: newborns afflicted with the condition tasted salty when kissed by their parents. This sweet and affectionate act has spurred advancements across various biomedical engineering fields, from diagnostic methods to molecular biology, as researchers delve into uncovering the disease’s underlying mechanisms and potential treatments.
One distinctive indicator of Cystic Fibrosis is the elevated negative potential observed in the sweat ducts of affected patients. This anomaly in membrane conductance is attributed to the impaired permeability of chloride ions. The primary cause of this impairment lies in mutations within the genes responsible for encoding a protein found in the transmembrane conductance regulator — a channel crucial for modulating chloride ion concentrations inside and outside cells. Specifically, this mutation disrupts molecular splicing patterns, resulting in the abnormal transcription of RNA strands, as illustrated in Figure 1 (A and B).
Point-of-Care Devices Stepped In
Point-of-care devices can access the intricate molecular information within sweat, and this extends beyond just chips to include wearable devices. For example, these wearable devices can decode the serotonin level hindered in the sweat, since the serotonin is a molecule significantly influenced by an individual’s emotional state. Thus, before delving into the realm of wearable devices, it is crucial to design a chip specifically tailored to extract targeted ion concentrations in the sweat of individuals with cystic fibrosis.
The fundamental design of this chip is illustrated in Figure 2, comprising three key segments: 1) Sensor Module, 2) Hardware Neural Network, and 3) Output Classification.
To selectively target the chloride anions in the sweat, both ion-selective electrodes and ion-selective organic chemical transistors (IS-OECT) are employed to stabilize the signal inputs linked to the neural network. The updated weight in the machine learning algorithm is regulated by the arrayed electrochemical random-access memory (ECRAM). Since the traditional organic electrochemical transistor coupled with ionic-electronic conductors cannot operate locally and differentiate patients’ personalized situations, this point of care device encodes the neural network reflected by the circuits, effectively returning the unbiased testing results to the on-site patients.
Specifically, after accessing the concentration of particles from the sweat, a threshold within the binary selection (high concentration or low concentration) is defined: chloride (high: -15mV, low: -45 mV) and potassium (high: -90mV, low: -60mV) respectively, and a bias section is maintained at the voltage of 60 mV. After the concentration of the particles being detected, the hardware neural network, which is composed of the feed-forward propagation and backpropagation, was initiated correspondingly to update the weights from the inputs of particle concentrations, and the weights while a positive diagnosis (The patients without the cystic fibrosis) is detected from the output classification (Fig2. (A)). [A setup of the off-set circuit was used to amplify the signals coming from the sensor. Additionally, the active circuit in the PCB is used to generate the voltage outputs that can quantify the particle concentration (Fig2. (B)).]
The tested results:
Therefore, this device sheds light to the futuristic neuromorphic platform that can encode machine learning at the hardware level to facilitate the diagnosis of cystic fibrosis disease.
Other Neuromorphic Networks
Besides, a more innovative design of the memristive device that mimics the biological neural network can be classified by the excitatory and inhibitory synapses in the input layer before the neuronal activation, and a bias section could be involved for the signal clearances and artifacts elimination. To process extensive datasets and save a certain amount of energy in the future, a multi-layer architecture at both software and hardware level, such as spiking neural networks (SNNs), is called to compare with the traditional artificial neural networks (ANNs). In addition, its application can be expanded to the blood test by combining with targeted bio-conjugates (antibody or fluorescent proteins) to promote a next-generation point-of-care diagnosis chip to unbiasedly reveal the spectrum of diseases.
References:
van Doremaele, E.R.W., Ji, X., Rivnay, J. et al. “A retrainable neuromorphic biosensor for on-chip learning and classification.” Nat Electronics, 2023, https://doi.org/10.1038/s41928-023-01020-z
Quinton, P. “Chloride impermeability in cystic fibrosis.” Nature, 1983, 301, 421–422, https://doi.org/10.1038/301421a0
Cutting, G. “Cystic fibrosis genetics: from molecular understanding to clinical application.” Nat Rev Genet, 2015, 16, 45–56, https://doi.org/10.1038/nrg3849
S Battistoni, V Erokhin, S Iannotta. “Organic memristive devices for perceptron applications, S Battistoni.” Journal of Physics D Applied Physics, 2018, doi:10.1088/1361–6463/aac98f.
Quinton PM. “Physiological basis of cystic fibrosis: a historical perspective.” Physiol Rev, 1999, doi: 10.1152/physrev.1999.79.1.S3.
