ACAD system for automatic dysplasia grading on H&E cervical whole-slide images

Sep 24, 2026

Uploaded by moss~

A weakly-supervised computer-aided diagnosis (CAD) framework is proposed to grade cervical dysplasia directly from H&E cervical whole-slide images (WSI). The pipeline combines an epithelium segmentation step with a dysplasia classifier distinguishing non-neoplastic, LSIL, and HSIL, removing the need for manual epithelial region identification. Slide-level evaluation on 600 independent publicly available samples reports balanced accuracy of 71.07% and sensitivity of 72.18%, aiming to support objective pre-cancer detection for timely screening and treatment.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Cervical cancer pathology

Grading of cervical dysplasia (LSIL/HSIL)

Proposed CAD system

Weakly-supervised dysplasia grading on WSI

Epithelium segmentation and dysplasia classification

Dataset and performance evaluation

[www. nature.com/scientificreports](www. nature.com/scientificreports) OPEN ACAD system for automatic dysplasia grading on H&E cervical whole‑slide images Sara P. Oliveira1,2,5, Diana Montezuma3,4,5*, Ana Moreira2,5, Domingos Oliveira3, Pedro C. Neto 1,2, Ana Monteiro3, João Monteiro3, Liliana Ribeiro 3, Sofia Gonçalves3, Isabel M. Pinto 3 & Jaime S. Cardoso1,2 Cervical cancer is the fourth most common female cancer worldwide and the fourth leading cause of cancer‑related death in women. Nonetheless, it is also among the most successfully preventable and treatable types of cancer, provided it is early identified and properly managed. As such, the detection of pre‑cancerous lesions is crucial. These lesions are detected in the squamous epithelium of the uterine cervix and are graded as low‑ or high‑grade intraepithelial squamous lesions, known as LSILand HSIL, respectively. Due to their complex nature, this classification can become very subjective. Therefore, the development of machine learning models, particularly directly on whole‑slide images (WSI), can assist pathologists in this task. In this work, we propose a weakly‑supervised methodology for grading cervical dysplasia, using different levels of training supervision, in an effort to gather a bigger dataset without the need of having all samples fully annotated. The framework comprises an epithelium segmentation step followed by a dysplasia classifier (non‑neoplastic, LSIL, HSIL), making the slide assessment completely automatic, without the need for manual identification of epithelial areas. The proposed classification approach achieved a balanced accuracy of 71.07% and sensitivity of 72.18%, at the slide‑level testing on 600 independent samples, which are publicly available upon reasonable request. Cervix uteri cancer is the fourth most prevalent cancer, the fourth cause of cancer-related death in women, worldwide, and the ninth leading cause of female cancer, in Europe1. Around 80–90% of cervical cancers are squamous cell carcinomas (SCC) and the large majority of these are caused by human papillomavirus (HPV) infection, although HPV-independent forms of cancer also exist2. Fortunately, cervical cancer is one of the most successfully preventable and treatable forms of cancer, as long as it is detected early and effectively managed3. Thus, screening pre-cancerous lesions and vaccination are key to preventing the disease. Squamous intraepithelial lesions (SILs) of the uterine cervix, or cervical intraepithelial neoplasia (CIN), are pre-malignant HPV-driven proliferationsof the squamous epithelium, showing viral cytopathic changes and/or maturation alterations that do not extend beyond the basement membrane2. Importantly, the World Health Assembly adopted a global strategy for cervical cancer elimination in August 2020, which is set on 3 pillars: vaccination, screening and treatment4,5. Each country should meet the 90–70–90 targets, by 2030: 90% of girls fully vaccinated against HPV by the age of 15; 70% of women screened using a high-performance test by the age of 35, and again by the age of 45; 90% of women with pre-cancer treated and 90% of women with invasive cancer managed4,5. Thus, technological innovation (new and better tools) is needed to achieve cervical cancer screening and management targets, as stated by Rodriguez et al.6, as well as the strategic implementation of automatic diagnosis tools in clinical practice. In computational pathology, there are several examples of studies focused on invasive cancer detection7–9 and grading10–12, but in cervical pathology, detecting and grading intraepithelial neoplasia has paramount significance, as detection of the disease in a precancerous state allows for close monitoring and effective treatment, preventing the development of cancer. To this end, the main contributions of this work are (1) a computer-aided diagnosis (CAD) system for dysplasia grading on cervical whole-slide images (WSI), which includes (2

FAQs

What problem does the proposed ACAD system address?

It addresses the subjectivity of grading cervical dysplasia by enabling automatic LSIL/HSIL assessment from H&E whole-slide images.

How does the framework perform dysplasia grading?

It first segments squamous epithelium, then applies a weakly-supervised classifier to label slides as non-neoplastic, LSIL, or HSIL.

What performance results are reported and on what dataset?

On slide-level testing with 600 independent samples, it achieves balanced accuracy of 71.07% and sensitivity of 72.18%, using slides available upon reasonable request.

Sep 24, 2026

Uploaded by moss~