Abstract
In Taiwan, kidney disease is an usual disease and common issue. Peritoneal dialysis (PD) is common way to treat kidney disease patient. We have develop a portable homecare system to supervise the patients' dialysate condition and record for the doctors to diagnosis. However, our expectation on this system is to help the users can grasp their health condition immediately, and give them a suggestion about visiting the doctors or not. Therefore, my research focused on the develop of the algorithm which can preliminary screen out the inflammation samples and normal samples for doctors and users to refer. Different from the traditional biomedical testing, our system use colorimetric information and absorbance to screen out the peritonitis from patients with PD treatment. To collect the colorimetric of the PD effluent, we should pour the PD effluent into the transparent container, then take a picture for the color checker and the container with effluent. To get the absorbance of the PD effluent, we use arduino micro controller to drive the light emitter to incident the light pass through the PD effluent and calculate the light intensity to get the absorbance of the PD effluent. After these steps, the data will be sent to the cloud data base and we can analysis the patients' condition with our screening method base on the data we collected. In our screening method, we establish a three steps flow for the screening. First, using gaussian mixture model (GMM) cluster the samples of PD effluent into undetectable and detectable two groups. This step we take the colorimetric information from the dialysate for the features. Thus, the undetectable samples will be exclude. and the detectable samples will be sent to the next screening stage. Second, according to the hypothesis test, we calculate a standard threshold according to the turbidity data collected by National Taiwan University Hospital (NTUH) and we. Operating with this standard, according to the the region where the samples are distributed to, once their region are located on the fuzzy region, then the third step will be operated to solve it. In the last stage, we use the exponential moving average (EMA) to calculate the same users' historical data and compare with it to see whether the newest sample getting better or not. In brief, these three stages screening can give a simple and fast determination after the users finish the whole screening. So far, we have 75 samples, based on this database, this algorithm can reach 94.5\% accuracy. In summary we design an screening algorithm on our homecare system, and it not only improve the accuracy from the original system but also adopts the characteristic of the fast screening and easy to operate. We believe this system can bring the users a better life.