Escudo de la República de Colombia Escudo de la República de Colombia
UN Periódico Digital

Resultados de Búsqueda:

UN Periódico Digital
Artificial intelligence helps reduce time to diagnose breast cancer

The cooperation between pathologists, oncologists, computer and biomedical specialists is now improving early cancer diagnosis by integrating technology and computer methods, in a new area of science known as digital pathology, which optimizes traditional pathology.

One of the branches of biology is pathological histology or histopathology, which is the study of diseased tissue, which helps identify abnormal patterns and diagnose diseases such as cancer. Samples of these tissues or biopsies are placed on top of microscope slides for researchers to analyze.

While the human body and its organs are tridimensional, pathologists can only analyze tissue samples in two dimensions. For this, they use a microscope and observe a glass slide with a very thin slice of tissue, tinted with special dyes to highlight the cell nuclei and cytoplasms.

 “The appearance, normal or not, of a tissue depends on the organ, the place of the body where it is, the dyes used and the direction of the cross-section, besides the natural anatomical variability between people and the pathological abnormalities,” explains Universidad Nacional de Colombia (UNal) IT Department Ph.D. Ángel Alfonso Cruz Roa, who has contributed significantly to the field of digital pathology.

Aware of the importance of early disease detection for an opportune treatment, Cruz carried out his doctoral thesis project called Data-driven Representation Learning from Histopathology Image Databases to Support Digital Pathology Analysis. His goal was to contribute to the optimization of diagnostic support for different types of cancer, such as breast and skin cancer.

The innovation of his research project was to use digital histopathology virtual imagery to analyze diseased tissues instead of using traditional microscopic methods with glass slides.

According to the researcher, for this, they digitalized tissue samples with robotic slide scanners which help obtain virtual digital histopathology slides with great resolution and great storage capacity (100 megapixels and 20 GB). Therefore his work has optimized image management for diagnostic work, despite the size and complexity.

Deep Learning, an innovative method

In a joint project between UNal and Case Western Reserve University, researchers analyzed a great amount of patient data and digitalized images. “They are working in other types of diseases such as skin and brain cancer,” said Cruz.

According to the researcher, UNal has been working with deep learning, which consists of a set of automatic learning algorithms applied to skin cancer, along with researchers Dr. Fabio González and Eduardo Romero.

“In 2012, Dr. Anant Madabushi presented us the challenge of verifying if our method based on deep learning could work with their data. Therefore, we began to work with brain cancer in children, known as medulloblastoma. In this case, we used a deep learning method known as autoencoders neural networksand obtained better detection results because we improved the capability to distinguish between the types of tumors using our methods,” says the researcher.

Artificial intelligence uses a method of error learning, improving, minimizing and going forward. This allows the technique to correct and learn from its mistakes and learn to recognize patterns with the provided data. Similar to how the brain learns from external stimuli, establishing neural connections.

 “When I was in the final stages of my thesis project, there were still doubts of the regulations for use of virtual slides for diagnosis. However, in 2017 the Journal of the American Medical Association of Dermatologypublished a research project in which there was a concordance of 94% between the diagnosis using histology and microscopy slides and digital virtual histology images in 499 dermatological cases; therefore the diagnostics of digital cases is equivalent to traditional methods. This is a great opportunity which did not exist before and uses technological digital pathology advancements for the community,” Cruz.

His work carried out with Drs. Fabio González, Anant Madabushi, and Haibo Wang, already has a patent for automatic cancer tumor mitosis detection in the United States. Additionally, there is another patent application in course, in the context of automatic and efficient high-resolution histopathology analysis to detect invasive breast cancer tumors.

Early cancer detection continues to be one of the main issues to treat diseases, mainly in developing countries such as Colombia, despite the progress in science in the last decades. This is quite a concerning problem as discovering cancer opportunely is crucial for determining treatment and save 8,686 annual average cases in Colombia, according to the Colombian Ministry of Health.

In Colombia, pathology specialists are scarce and are generally located in the main cities, but patients can live wherever. Many times a sample can take up to a month to process, added to the challenge of working with low bandwidths existent in Colombia.

The thesis research project was deemed outstanding.

Read the complete theses project here (in Spanish)


Consejo Editorial