Computational Pathology for Breast Cancer and Gynecologic Cancer

Advances in computation pathology have continued at an impressive pace in recent years [...].

(ERS) from slides that have been H&E-stained. Similarly, on biomarker status prediction, Shamai et al. [6] showed that a convolutional neural network-based framework of wholeslide images (H&E-stained) can predict the expression of specific biomarkers (e.g., PD-L1). The authors of this study make use of a large tissue micro-array repository that contained various respective stains for different biomarkers (e.g., IHC for PD-L1) and a collection of H&E-stained whole-slide images. Additionally, a highly experienced pathologist annotated the expression of PD-L1 for each sample in the datasets.
On the other hand, the World Health Organization (WHO) estimated that cervical cancer affects approximately 300,000 individuals annually, with less developed nations contributing to more than 85% of these deaths in recent years. Cervical cancer, one of today's most prevalent cancers in women's health, is diagnosed in one woman per minute [13]. The cervical cancer dataset has been used in various types of research to analyze tissue samples for the study of the disease. Wang et al. [14] introduced a modified FCN for cervical HSILs or higher (SQCC) segmentation of WSIs using Pap smear specimens for use in real-world situations and discovered that the proposed cervical Pap smear diagnosis method could aid with the automatic recognition and quantification of cervical HSILs or higher (SQCC). Zhu et al. [15] proposed an assistive diagnostic, AIATBS, to optimize the assessment of cervical cytological smears using clinical Bethesda system (TBS) standards. Cheng et al. [16] developed a three-stage DL-based model, consisting of a low-resolution model to locate lesion areas at the low-resolution setting, a high-resolution model to recognize the top 10 lesion cells to be fed to the slide-level classification model, and recurrent neural network (RNN)-based technique to integrate the 10 most suspicious input features and yield the result for slide-level prediction.
In addition, ovarian cancer is the second highest frequent gynecologic cancer among women globally [17]. Over 140,000 women die from ovarian cancer worldwide every year, and over 225,000 women are diagnosed with it [18]. To discover more about computational pathology in ovarian cancer, some experiments were conducted utilizing ovarian cancer datasets. Wang et al. [19] proposed a modified FCN technique for predicting the therapeutic efficacy of bevacizumab on patients with ovarian cancer based on histopathological (H&Estained) whole-slide images, without using any pathologist-provided regionally annotated areas. In a further study, Wang et al. [20] built a modified FCN-based precision oncology system using immunostained tissue microarray (TMA) whole-slide images to determine bevacizumab therapeutic impact in patients with epithelial ovarian cancer (EOC) and peritoneal serous papillary carcinoma (PSPC). In another computational pathology application, Boehm et al. [21] compiled a multi-modality dataset of 444 ovarian cancer patients with the majority of the high-grade serous carcinoma type and found quantitative characteristics on multimodal imaging affiliated with prognosis (e.g., tumor nuclear dimension on H&E staining and omental texture on contrast-enhanced computed tomography).
An analysis of histopathology images involves more than just visual analysis; it also involves data integration from patient clinical information and demographic details [22,23].
Most clinical data appears in unstructured free-text reports, including demographic information, patient test results, patient medical history, and clinical outcomes. The AI-based system role will be crucial in sorting through these many sources of information and supporting pathologists in making the best treatment decisions for patients. The AI-based medication must incorporate clinical data as well as histopathology images, allowing for assessment that exceeds the capabilities of the human brain alone. Rich data resources will aid pathology's transition from a clinical to an informatics science, with tissue images supplied as one of the information sources [2].
The approaches for AI-based image recognition and its combination with other information sources have developed well within the area of computational pathology toward the level where they might eventually be ready to be converted from the research environment into real clinical laboratories. Despite computational pathology offering a lot of potential avenues, there are still some challenges to overcome [24]. Therefore, we are issuing a call for research papers that utilize innovative approaches for creating future studies to address current challenges in computational pathology for breast cancer and gynecologic cancer. Readers can find more information about publishing research in Cancers in the Special Issue "Computational Pathology for Breast Cancer and Gynecologic Cancer".
Funding: This research study is supported by the National Science and Technology Council, Taiwan (MOST 109-2221-E-011-018-MY3).

Conflicts of Interest:
The authors declare no conflict of interest.