Researchers at the Kwame Nkrumah University of Science and Technology (KNUST), Kumasi, in collaboration with partner institutions, have developed a promising artificial intelligence (AI) model that could significantly improve the accuracy and accessibility of breast cancer diagnosis.

This innovation is crucial for healthcare systems in low-resource settings where medical specialists and diagnostic infrastructure may be limited.

Led by Albert Dede, a doctoral candidate, the research addresses a critical challenge in medical imaging: how to effectively analyze large, high resolution histopathology images used in cancer detection.

Histopathology slides are microscope images of tissue samples that play a vital role in cancer diagnosis.

However, due to their size and complexity, these images are difficult for both human experts and traditional AI systems to interpret accurately. Important diagnostic features, such as the subtle shape and size variations of cancer cells, can easily be overlooked.

To overcome this, the researchers combined two advanced AI techniques. The first, wavelet analysis, breaks down an image into smaller, more manageable segments while preserving key details.

The second, deformable convolutions, enables the model to adapt to variations in cell shapes and structures, allowing for a more flexible and responsive analysis similar to how a trained pathologist might visually scan a slide and focus on areas of concern.

The AI model was tested using the BreaKHis dataset, a widely used benchmark in breast cancer image classification. It achieved a diagnostic accuracy of 96.47 percent at the image level and 96.55 percent at the patient level. Notably, the model performed exceptionally well when analyzing images captured at high magnification (200×) where fine tissue details are most critical.

An important feature of the model is its relatively low computational demand, which makes it more feasible for use in hospitals and diagnostic centers that lack high end computing resources or expert personnel.

“This approach not only improves diagnostic accuracy but also makes the technology more accessible to under-resourced healthcare environments,” the researcher noted.

The study, titled "Wavelet Enhanced Deformable Convolutional Network for Breast Cancer Classification in High Resolution Histopathology Images," is published in the journal Applied Intelligence.