Image classification of unlabeled malaria parasites in red blood cells

Abstract

This paper presents a method to detect unlabeled malaria parasites in red blood cells. The current "gold standard" for malaria diagnosis is microscopic examination of thick blood smear, a time consuming process requiring extensive training. Our goal is to develop an automate process to identify malaria infected red blood cells. Major issues in automated analysis of microscopy images of unstained blood smears include overlapping cells and oddly shaped cells. Our approach creates robust templates to detect infected and uninfected red cells. Histogram of Oriented Gradients (HOGs) features are extracted from templates and used to train a classifier offline. Next, the ViolaJones object detection framework is applied to detect infected and uninfected red cells and the image background. Results show our approach out-performs classification approaches with PCA features by 50% and cell detection algorithms applying Hough transforms by 24%. Majority of related work are designed to automatically detect stained parasites in blood smears where the cells are fixed. Although it is more challenging to design algorithms for unstained parasites, our methods will allow analysis of parasite progression in live cells under different drug treatments.

A Systematic Review on Automatic Detection of Plasmodium Parasite

“…Preedanan's study [27] collected 15 images with the dimensions of 1360 × 1024 pixels. Zhang's study [28] used 43 images, and Rosado's study [29] used 194 images. After that,…”

Artificial Intelligence in Malaria Diagnostics

“…Shan Yang 1, 2 , Yanbing Li , Tuoyu Liu 1, , Haidong Zhang 4 区，RDTs 比血涂片镜检的成本更高 [14] 。因此，目前显微镜检和 RDTs 更多的是相互补充， 而不是相互取代。 与上述两种方法相比，PCR 分子检测方法具有更高的灵敏度和特异性 [7] ，是所有测试 中最准确的方法。PCR 可以检测血液中极低浓度的疟原虫，并能鉴别疟原虫种类。然而， PCR 技术复杂且成本较高，其检测需要经过培训人员进行操作，检测过程需数小时。根据 Tangpukdee 等研究 [7] Otsu 阈值 [15][16][17][18][19][20][21][22][23][24][25][26] 基于图像包含双峰直方图的假设计算最佳阈值 (自适应)直方图阈值 [21,[27][28][29][30][31][32][33][34][35][36][37][38] 难以确定阈值…”

“…Hough 变换 [27,28,37] 需测定血细胞弧度，包括半径、形状 K-均值聚类 [46] 无监督学习技术，使用像素特征描述符通过迭代将像素 分配给 K 聚类 分水岭算法 [22] 提取连续边界区域，但经常出现过度分割 标记控制分水岭算法 [15,23,38,[47][48][49][50][51] 主要用于分割相互接触的细胞 主动轮廓模型 [47,49,[52][53][54] 基于水平集，确保拓扑灵活性，计算量大 基于规则的分割 [55] 需要了解细胞形状、大小、颜色等。 基于模糊规则的分割 [56] 当不确定性较高时，制定规则较困难 模糊发散分割 [57] 神经网络 [58] 需要判别能力较强的特征来区分前景和背景像素 模板匹配 [59] 自适应高斯混合模型距离变换 [60] 距离转换 [61] 自适应提升算法 [62] 查找表 [63] 标准化切割算法 [64] 计算量大 表 2 厚血涂片分割算法 Otsu 阈值 [65] 基于图像可分为两类且遵循双峰直方图的假设，计算最佳阈值 直方图阈值 [44,66,67] 难以确定阈值，通常与其它方法并用以提高性能 形态学运算 [68] 数学形态学运算，包括粒度运算、开运算、闭运算等。用于表征血细胞的弧状外形、 大小、边界、骨架、纹理、渐变等。 模糊 C-均值 [39] 2.…”

Implementation of Siamese Convolutional Neural Network from Cell Images for Malaria Disease Identification

“…Microscopic examination is a technique that often used to detect malaria parasites [3]. Microscopic examination has some limitations, this examination needs a high-quality microscope and it needs an expert who has experienced in using a microscope.…”