A fractal based approach to evaluate the progression of esophageal squamous cell dysplasia. (February 2019)

• Record Type: Journal Article
• Authors: Hosseini, Zahra Sadat; Hashemi Gholpayeghani, Seyed Mohammad Reza; Sotoudeh, Masoud; Malekzadeh, Reza
• Is Part Of: Biomedical signal processing and control. Volume 48(2019)
• Page Start: 273
• Page End: 289
• Publication Date: 2019-02
• Subjects: Fractal analysis -- Self-similarity -- Classification -- Squamous cell dysplasia -- Endoscopic biopsy image; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2018.09.001
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A new algorithm for detection of early stage of esophageal squamous cell dysplasia based on Vienna grading system. Applying the fractal analysis to characterize the architectural and cytological properties of esophageal epithelial tissue. Propose a new method to calculate the fractal dimension. Suggest low-cost adjunctive information to classify borderline lesions that cannot be determined easily by the visual system. A first computer-aided system to eliminate discrepancies and subjectivities in the diagnosis of esophageal LGD. Abstract: Esophageal squamous cell carcinoma (ESCC) is the most prevalent malignancy of the esophagus with a very poor prognosis. Nevertheless, squamous cell dysplasia (ESD) has been identified as the only histological precursors of ESCC. Since, tissue alterations are slight in the early stage of ESD, human diagnosis is subjective. Hence, this work presents a first computer-aided system to differentiate low-grade dysplasia (LGD) from normal esophageal mucosa according to Vienna grading system, which is the most widespread method for histological grading of esophagus tissues. We captured microscopic images of a well-oriented region of Normal and LGD biopsies to characterize the architectural and cytological properties of specimens based on the computational analysis. We produced two sets of enhanced images. Then, by considering the fractal concept, we defined a new scale-dependent function in the generalized fractal dimension formulation to include the special information of both preprocessed images together. Then, for each image, a pattern was computed from variations of tissue fractal geometry across the pathway of dysplasia development. We proposed features extracted from these patterns to classify deviations of tissue characteristics from the normal stage. This method successfully differentiated the two diagnosis classes with statistical significance and high performance (accuracy = 97.78% ± 0.05, p < 0.0001). To approve the self-similar property of the esophagus tissue and evaluate the robustness of this technique, it was conducted at two image magnifications and repeated for different biopsy sizes. Our results confirm that this tissue is a multifractal object and fractal analysis effectively extends the conventional light microscopy method allowing for an early detection of ESD. Thus, computer-aided detection can support pathologists' diagnosis and result in a consistent decision. On the other hand, generally, the proposed method can be used to estimate the fractal dimension of other images.