Image processing methods for the structural detection and gradation of placental villi
Introduction
Spontaneous miscarriage is the pregnancy that ends on its own within the first 20 weeks of gestation. The average risk of miscarrying is between 10% and 25% [1], [2], [3], [4], and about 80% of miscarriages happen in the first trimester. The causes of spontaneous miscarriage are poorly understood. A woman who has had a previous miscarriage has a 25% chance of having another. A woman with two previous miscarriages has a 28% risk of another miscarriage, and for three previous miscarriages, the risk increases to 43% [1], [2], [3], [4]. Therefore, correct diagnosis as well as effective treatment are important.
The analysis of the placental structures in spontaneous miscarriages (in the first trimesters of pregnancy) is complex [5], [6], [7], and it involves qualitative histological assessment of placental tissue under the microscope. The placenta is formed and it keeps changing during pregnancy. Abnormalities in placental structure and function may be critical for the fetus, and can lead to spontaneous miscarriage [8]. Microscopic analysis of the placenta can provide valuable information related to the causes of the spontaneous miscarriage [9]. Placental pathologies are associated with infarction (25% of cases), chorioamnionitis (12.5% of cases), molar changes (9% of cases), hypervascularization and avascularization according to [10]. The latest methods for qualitative and quantitative assessments of placenta from miscarriages include: determining the presence of villi, counting of vessels inside of villi, and evaluation of villi edema. According to [9], [10]:
- •
If villi are not established, the fetus cannot develop properly;
- •
The insufficient number of blood vessels (avascularization) can be an effect of inadequate angiogenesis of the fetus;
- •
Too many blood vessels (hypervascularization) can result from the disorders in the osmotic transfer between the fetus and the mother;
- •
The edema (storage of excessive fluid inside the villi) is a response to the hydropic problem;
- •
The occurrence of hemorrhages on the images of placenta suggests coagulation-related disorders.
The histological evaluation of the placental specimen provides information that helps to identify the causes of spontaneous miscarriages and it may subsequently help to avoid the next miscarriage.
The complexity of placental tissue is difficult to assess manually. In the literature, several studies propose methods for the automatic recognition of villi structures in the placenta [11], [12], [13], [14], [15], [16], [17]. The aim of the research performed in Refs. [12], [13], [14] was the radiological evaluation of placenta during pregnancy. Alansary et al. [13] have presented methods to delineate the human placenta in MRI scans. Stevenson et al. [14] have described a method for placenta segmentation in 3D ultrasound images, while Ayache et al. [12] have proposed a method for automatic recognition of placental tissues and staging of its development. Placental development was assigned to four classes, based on features extracted by discrete wavelet transform (DWT) and further classified by a multilayer perceptron (MLP). Under the microscope, the villi are evaluated based on the number of blood vessels and the degree of edema. Almoussa et al. [11] have presented a method of automatic detection and extraction of blood vessels from a histological image of placenta by using image processing techniques and neural networks. In our previous work [15], [16], [17], we have presented algorithms that can recognize individual structures (villi, trophoblast, and vessels) on the histopathological images of placenta. In this work, we have presented an automatic system for the villi detection and evaluation (based on edema degree and vessels number). To the best of our knowledge, no existing solutions can automatically support the pathological diagnosis of spontaneous miscarriages.
The aim of present work is to evaluate villi after spontaneous miscarriage by using microscopic image analysis and to supplement the manual slide reading with a quantitative readout. To the best of our knowledge, no work presenting the automatic system of recognition and evaluation of placental structures has been reported until now. The novelty of the presented work is the automatic detection and analysis of placental villi on the images, as well as the automatic evaluation of edema degree.
The proposed algorithm is based on texture techniques [19], [20], color normalization [18], [21], mathematical morphology [22], and region growing [23]. Our method that can segment out villi, villi edema, villous mesenchyme, trophoblast, collagen, and villi vessels were quantitatively evaluated using manual delineation of these tissue components by a pathologist. The results have been compared with the medical expert annotations.
In Section 2 we describe the materials used in this study, in section 3 we present the proposed methods and algorithms. The summarized results and conclusions are shown in sections 4 Results, 5 Discussion, respectively.
Section snippets
Materials
Thirteen histological slides of the placenta stained with hematoxylin and eosin (H&E) were retrieved from the archives of the Department of Pathology, Military Institute of Medicine, Warsaw, Poland. The slides were digitized using 3DHistech Pannoramic 250 Flash II scanner with 20 × lens magnification—the effective resolution was 0.389 μm per pixel. Digital slides of whole slides (WSI) were encoded as 24 bit RGB matrices and saved in mrxs format. The data of the patient were removed from glass
Methods
The presented method consists of three major steps: preprocessing, tissue detection and background removal, and quantitative analysis. Fig. 2 illustrates the general architecture of our approach. In the first step, the image is preprocessed for the segmentation and quantitative analysis. In the second step, we differentiate between various areas such as villi, collagen, villous mesenchyme, edema area, and blood vessels. The quantitative analysis includes counting of blood vessels and measuring
Placenta structure detection
The first experiment focused on the detection of different structures present within the placenta (Fig. 12). Using the method presented above, the objects such as villi, collagen, trophoblast, villous mesenchyme, and blood vessels have been recognized.
For the evaluation, 50 villi from different images were randomly selected by the medical expert. Only the coherent villi, wholly visible on the image and without mechanical damages or artifacts were selected for the analysis. In order to evaluate
Discussion
In the placental structure detection experiment, mesenchyme and villi structures were correctly detected. The villous mesenchyme detection compared to villi detection was less precise due to the noncoherent and ambiguous border between the trophoblast and the villous mesenchyme. Therefore, the modest underestimation occurred.
In the second experiment, the automatic evaluation of villi edema degree was 84%, and the automatic evaluation of the number of vessels was 86%. In two cases, the
Conclusions
The automatic approach to the segmentation and classification of complex tissue structures on histopathological images of placenta was presented in this study. We have developed a novel approach based on texture technique, mathematical morphology, and region growing. The normalization operation allowed reducing the influence of staining intensity in the structure detection process. The proposed method was used to recognize the internal structures of the placental tissue obtained after
Funding sources
This work was supported by the National Science Council, Poland, grant number 2012/07/B/ST7/01203.
Conflict of interest statement
The authors report no conflict of interest.
References (26)
- et al.
3-D Ultrasound segmentation of the placenta using the random walker algorithm: reliability and agreement
Ultrasound Med. Biol.
(2015) - et al.
Machine learning approaches to analyze histological images of tissues from radical prostatectomies
Comput. Med. Imaging Graph
(2015 Dec) - et al.
A comparative study of texture measures with classification based on feature distributions
Pattern Recognit.
(1996) - et al.
OpenSlide: a vendor-neutral software foundation for digital pathology
J. Pathology Inf.
(2013) - et al.
MIAP–Web-based platform for the computer analysis of microscopic images to support the pathological diagnosis
Biocybern. Biomed. Eng.
(2016) - et al.
Incidence of early loss of pregnancy
N. Engl. J. Med. 28
(1988) - et al.
Conception to ongoing pregnancy: the ‘black box’ of early pregnancy loss
Hum. Reprod. Update
(2002) - et al.
The Pathology of the Human Placenta
(2006)