Astri Handayani, ST, MT
Development of Machine Learning Techniques for Diabetic Retinopathy Classification (Astri Handayani) Diabetic Retinopathy Classification Using Machine Learning Techniques has been widely developed but with different levels of accuracy caused by differences in the quality of data used in each study and disinancement of data proportions for each retinopathy diabetic class. The difference in the level of accuracy is found in the classification of proliferative class Diabetic Retinopathy (PDR) which generally has a representation of the data samples of the most highly high intra-class diversity. Patients with PDR are very likely to experience intra-retinal bleeding which often leads to permanent blindness. Therefore, this study examines the ability of various models of machine learning that has existed in identifying PDR and investigating different approaches to optimizing the capabilities of the Machine Learning model in detecting PDR cases. This study compiles various public and private datasets that provide data on PDR data, as well as several Machine Learning models that have the best performance in detecting diabetic retinopathy. Performance of different engine learning models compared to compiled PDR data samples, then optimizations for each of these models. The optimization strategy includes the transfer of learning, the addition of cascaded linear classifiers after CNN to optimize the distinction of PDR and NPDR, as well as the use of spatial features extraction techniques to localize the retinal area where the proliferation of blood vessel growth (neovascularization) occurs. Improved Performance of Fetal Head Measurement Algorithms on Ultrucous Examination (Astri Handayani) This study focuses on developing a denoising system to improve the image quality of ultrasound and optimization of the fetal head measurement algorithm so that it can provide better or comparable accuracy with existing but more efficient in terms of processing time. This study uses several median-based filter-based techniques tried in the Speckle Noise simulation in the ultrasound image and techniques that have optimal performance used in the image of fetal ultrasound. The pixel classification system is built to localize the fetal head candidate which is subsequently customized systems for early trimester data and end of pregnancy. Based on the identified fetal head candidate, an ellipse fitting process is carried out to calculate the fetal head circumference The main problem with the acquisition of skin image is the quality of imagery that depends on environmental conditions, such as lighting and skin color. This can be above using image processing, Image Enhancement. One method of Image Enhancement commonly used is a stretching histogram based. This research was conducted with exploration of Image Enhancement using the Contrast-Limited Adaptive Histogram Equalization (Clahe) and Multiscale Retinex Color Restoration (MSRCR) in skin cancer detection. The main objective of this research is to compare the effects of contrast to contrast using Clahe and MSRCR in early detection of skin cancer using CNN-based machine learning. The main limitations of this study are only two classification classes discussed, namely benign and malignant. Activities carried out in this activity include; Literature Studies, Implementation of Image Processing Engineering, Image Enhancement In the image of skin cancer by using Clahe and MSRCR, the manufacture of CNN-based Machine Learning models, data analysis and paper writing.
Proceedings of the International Conference.
Automatization of clinical extraction with the approach of learning machine has a broad impact on medical specialization, where; Early detection of diseases needs to be done on a large number of populations and with recurring periods, Early detection of illness is very dependent on qualitative or semi-quantitative doctors and subjective perceptions of patients, The number of specialist doctors is limited or distributed unevenly.