Artificial intelligence (AI) has gradually transformed healthcare over the last few years.Exhibition by Bohr & Memalzadeh1By enabling more accurate disease detection, image analysis, patient monitoring, and more efficient medication self-administration, it can impact every area of healthcare.2. Combining AI and health monitoring devices can significantly reduce medical costs1. Deep learning (DL), in particular, helps find hidden correlations and patterns using advanced machine learning algorithms such as artificial neural networks (ANNs).3,4. This allows us to use deep machine learning to recognize complex patterns in patient data earlier, detect anomalies, and correlate symptoms with disease.This new field of medicine could become more accessible and affordable5,6. By leveraging machine learning, a combination of low-cost, low-accuracy sensors could potentially achieve accuracy on par with state-of-the-art medical technology, reducing costs and providing more universal healthcare access. I’m trying to demonstrate that I can.1. Building on this philosophy, this report establishes how deep learning approaches can yield more accurate data predictions from ultra-low-cost temperature sensor arrays.
Temperature sensor
Temperature sensors come in a variety of designs and materials depending on the intended application. Beyond cost, important features to consider are reliability, response time, accuracy, sensitivity, temperature range and, in the case of skin temperature, wearability.7. They include thermocouples, resistance temperature detectors (RTDs), thermistors, semiconductor sensors, etc. Each has its own strengths and weaknesses.8, 9, 10. Detailed specifications of these sensors can be found in the literature.7,9. This project uses a negative temperature coefficient (NTC) thermistor and a semiconductor-based integrated circuit (IC). The NTC sensor measures changes in resistance. The temperature is then calculated using the Steinhart-Hart equation given as11,
$$\begin{align} \frac{1}{T} = A + B\ln {R} + C(\ln {R})^{3} \end{align}$$
(1)
where T is the temperature in Kelvin, R is the resistance of the thermistor, and A, B, and C are constants specific to the sensor device, usually provided by the manufacturer.11. These are analog sensors whose output is a continuous electrical signal that is converted to temperature. In contrast, integrated circuit (IC)-based sensors use bipolar transistors to make measurements. The specific IC sensor chosen for this work also contains an analog-to-digital converter. Therefore, the output signal from the sensor will be discontinuous temperature readings.
temperature in medicine
Body temperature is one of the important vital signs in health assessment. There are individual differences in normal temperature, but if it is normal, it is considered normal. \(37\;^\circ \hbox {C}\)12. However, the temperature will vary depending on the body part being measured, the ambient temperature, and the subject’s activity.ends get cold easily13. The human body has a thermoregulatory mechanism to keep the body temperature constant. \(37\;\;^\circ \hbox {C}\). These mechanisms can be activated by cold, such as shivering, hunger, and goosebumps, or by heat, such as perspiration and enhanced respiration.13. Temperature measurements are therefore particularly useful for detecting infections and inflammation, but more generally for detecting immune responses. Bacterial secretions and viral load cause fever, which can also be detected by an increase in body temperature.12, 14. Wearable devices have made it possible to continuously monitor several health indicators and vital signs outside the doctor’s office.14,15. For example, thermal imaging can be used to validate blood flow and detect small temperature changes in patients with vascular disorders.16. Another use is to monitor the course of illness and treatment through temperature, such as in pneumonia patients.17 or high-risk diabetics with foot ulcers18. Furthermore, temperature can not only be used to distinguish between superficial and deep skin burns, but can also be used to monitor and predict skin regeneration and healing processes.19.Also used to monitor infected wounds20,21. Thermal imaging is currently used almost exclusively in clinical settings. As a simple, non-invasive approach, it has many advantages, but its main drawback is that it works directly on the skin.17. This can lead to accessibility, comfort and privacy issues depending on the areas affected. Moreover, it can only be performed in medical settings. In contrast, temperature sensors can be easily implemented in wearable devices such as watches.14,patchtwenty two and even face maskstwenty three It’s even more hassle-free and allows for continuous patient monitoring. This can lead to calibration and accuracy issues, but we aim to improve wearable devices with DL models and learn how to compensate for these inherent issues.
machine learning
Accuracy and reduced accuracy of low-cost printed sensors are usually attributed to unreliable designs and cheaper materials and manufacturing techniques. An interesting idea, therefore, is to leverage machine learning algorithms on a statistically significant number of low-cost sensors (arrays) to potentially learn how to compensate for design and manufacturing flaws. So far, only a handful of research groups have reported significant progress in this exciting new approach to further improve the performance of inkjet-printed sensors beyond their physical limits. In 2019, he combined an array of 20 printed sensors with his two-step machine learning approach to produce an artificial nose used to classify foods.twenty four. There, a characterized random forest or k-nearest neighbor classification (with similar accuracy) is first used to classify into food categories (e.g. cheese, liquor, oil). A combination selector scan is then used to classify specific foods within that category (e.g. rum, vodka, whiskey, gin, tequila as liquor)twenty four. In 2020, researchers achieved sign-to-speech translation using a machine-learning-assisted stretchable sensor array.twenty five.On the other hand, biomolecular and protein sensing26,27can also detect gas and pollutant mixtures28,29 Demonstrations have also been made using low-cost printed sensors with machine learning processing. In 2020, a multidisciplinary team developed a smartphone-based malaria DNA diagnostic tool used in rural Uganda to improve connectivity between rural Uganda and intensive care facilities.30. They used a low-cost, paper-based, microfluidic diagnostic test and achieved over 98% disease detection accuracy.30. In all these cases, the final idea was to equip these low-cost sensor arrays with some intelligence to perform a specific function. classification task24, 25, 26, 27, 28, 29, 31.
Low cost sensors usually give poor and inaccurate results due to the quality of the equipment. This study shows that using a low-cost temperature sensor in combination with a deep machine learning framework improves accuracy and precision. This opens up a wide range of applications, especially in the medical field, where cheaper sensors can be placed on different parts of the body while predicting body temperature.
