Cervical Cancer Detection Accelerated with AI at CIC

Reporter: Adda Avendaño / Photographer: Javier González – August 14, 2026

Deep learning tools are being developed to support early diagnosis of this disease through the automated analysis of cytological images

Cervical cancer is one of the major public health challenges in Mexico. In the country, this neoplasm is the second leading cause of cancer-related death among women, after breast cancer, making early diagnosis essential to reducing its impact.

Cervical cytology, commonly known as the Pap smear, is the main screening method used in the public health sector, but it presents several challenges. It requires an experienced pathologist to interpret the results, involves long hours of work to analyze large numbers of samples—at least 50 slides under the microscope per day—and can sometimes lead to discrepancies between specialists.

To help reduce these processing times and contribute to faster diagnosis, Irari Jiménez López, a doctoral student in Computer Science at the Center for Computing Research (CIC), is developing Artificial Intelligence (AI)-based tools for the automated analysis of cytological images.

Mathematical Color Models

According to Irari Jiménez, machine learning and deep learning algorithms have increasingly been explored to detect abnormalities in cervical cell analysis. However, cytological samples contain images with numerous elements, creating a challenge for automated analysis.

To address this, the researcher developed a computational program that analyzes images of isolated cells using different color spaces. This strategy reorganizes the visual information contained in an image to highlight characteristics that can improve automatic cell classification while reducing the influence of less relevant features.

Jiménez López explained that color spaces are mathematical models used to represent colors through coordinate systems and noted that several electronic devices already rely on them.

“LED displays use the RGB system, based on red, green and blue, which can be combined to produce any other color. Printers, for example, use the CMYK system, whose magenta, yellow, cyan and black inks generate colors,” she explained.

Key fact

Persistent infections caused by Human Papillomavirus (HPV) are the most common cause of cervical cancer. They can cause morphological changes in cells, such as an enlarged nucleus or alterations in the cytoplasm.

The researcher explained that this process forms part of the image preprocessing stage. To determine which representation produced the best results for automatic cell classification, the team evaluated different color models. Twelve experiments were conducted, including image rotations and resizing to standardize the samples, providing the model with a broader range of examples and facilitating pattern recognition.

“We sought to study different color-space models that could help improve classification metrics, so we compared twelve different transformations, including RGB, CMYK, HSV, grayscale, CIELAB and YUV as image preprocessing techniques,” she noted.

Jiménez López explained that most digital devices and televisions represent images using the RGB color model. As an alternative, the research also evaluated the CIELAB color space, which was designed to approximate the way the human eye perceives color. The researchers hypothesized that this model could facilitate the identification of cellular structures and improve the performance of classification models.

Image Analysis with Deep Learning

In the next phase of the research, conducted under the supervision of CIC professors Carlos Aguilar Ibáñez and José Eduardo Valdez Rodríguez, the student developed a deep learning architecture—one of the most advanced areas of Artificial Intelligence—using computer vision techniques to analyze the images.

She explained that deep learning models are particularly useful for image analysis because they can identify patterns and spatial relationships between pixels while automatically extracting relevant characteristics that can subsequently be used for classification.

The researcher, who also holds a master's degree in Electrical Engineering, explained that the project uses supervised learning. During training, the model must be provided with the category corresponding to each image so that it can learn the relationship between the characteristics extracted from the images and their assigned labels, including cellular structures, morphological alterations, and quantitative analyses.

“Based on image preprocessing, we built a deep learning architecture to classify isolated cells. It includes a convolutional layer that extracts image features, a pooling layer that reduces their spatial dimensions, and an activation layer that determines which information is relevant and enables the neural network to learn the complex patterns it is provided with,” the researcher explained.

In addition, Irari Jiménez said the project includes the development of a segmentation algorithm designed to identify regions of interest within an image and distinguish cellular structures such as the nucleus and cytoplasm. This information makes it possible to analyze morphological and textural alterations, perform quantitative image analysis, and determine whether a cell is healthy or abnormal.

“The results obtained so far show accuracy levels above 95 percent in identifying isolated cells, a figure that demonstrates the potential of these tools to support the analysis of medical images,” said the CIC researcher.

The study has already been published in the international academic journal AI Magazine, a monthly publication by MDPI that is ranked in the first quartile (Q1) in the “Computer Science, Interdisciplinary Applications” category and is included in major academic evaluation indexes such as Scopus (Elsevier).

From Cell Analysis to Complete Samples

After achieving successful results in the classification of individual cells, the project has moved toward a more complex stage: analyzing complete images containing multiple cellular elements.

“This new phase of the research focuses on improving semantic segmentation, a computer vision technique that assigns a category to each pixel in an image. This will allow the algorithm to identify specific regions, such as cell nuclei, separate them from the rest of the image and analyze them individually,” the researcher explained.

The ultimate goal is to develop a system capable of examining a cytological sample in a manner similar to a pathologist working with a microscope: identifying the cells present, classifying them, automatically counting them and detecting possible abnormalities or the presence of other organisms, such as bacteria and fungi.

Although the technology is still under development, the research has already attracted the attention of medical specialists, who consider that this type of tool could become an important aid in clinical practice.

“The goal will never be to replace the work of healthcare professionals, but to provide them with auxiliary tools that can optimize analysis time, facilitate decision-making and expand diagnostic capabilities in institutions that serve large numbers of patients,” she emphasized.

Jiménez López stressed the importance of promoting interdisciplinary research that brings together engineering, computer science and medicine, since collaboration among specialists from different fields is essential for developing technologies that address real-world problems and generate tangible benefits for society.

For the doctoral student, the research also has a deeply human dimension. Her interest in combining engineering with health sciences led her to develop a project aimed at improving healthcare for women and helping ensure that the diagnosis and treatment of potentially life-threatening diseases can take place at earlier stages.