Application of Artificial Classification Techniques to Assess the Anemia Conditions via Palpebral Conjunctiva Colormap

• Main Authors: Ching-Chao Fan, 范景超
• Other Authors: Hwai-Tsu Hu
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/63416604728901576331

碩士 === 國立宜蘭大學 === 電機資訊學院碩士在職專班 === 101 === This study aims at developing a feasible approach to predicting the anemia conditions of people through the analysis of conjunctival congestion levels on eyelids, which can be used as a preliminary screen. The procedural steps include collecting eyelid conjunctiva photos of subjects, feeding extracted parameters into a personal computer, emulating expert interpretation of hemoglobin values, and learning rules for judging anemia conditions. In practice, photo images are taken using common digital cameras. Our experiments start with capturing an eyelid conjunctiva area on the photo image followed by extracting RGB values and luminance information pertaining to the specific area. Two techniques, namely neural networks and support vector machines, are employed to explore the relationship between the extracted parameters and real hemoglobin values. In such a manner, we can resort machine intelligence to predict the anemia condition. Amongst 103 cases for data analysis, 60 were used to train and 43 for test. The results obtained by neural networks indicate that the sensitivity and specificity can reach 0.6538 and 0.9412 respectively, of which the accuracy is almost same as professional judgment. Moreover, the likelihood ratio (+) is 11.1154, suggesting a high degree of reliability. On the other hand, the sensitivity and specificity achieved by support vector machines are 0.9615 and 0.8, respectively. The likelihood ratio (+) is 5.4487, which represents a moderate degree of reliability. In contrast, the likelihood ratio (-) is 0.0467, which belongs to a high degree of reliability. Based on our experimental results, it is concluded that the most reliable way to predict the anemia is the use of neural networks to analyze RGB color spectrum data. However, the exclusion of anemia can be better achieved by applying support vector machines to analyze red color spectrum data. Our study confirms that the eye conjunctival pallor degree can be adopted as a quick screening method for anemia. Using smart phones or digital cameras to capture the images for visual examination is a very convenient way to sieve out the anemia at a preliminary stage.