Morphological and textural descriptors analysis using artificial neural networks in mammograms

What is it about?

Mammograms BI-RADS 4 and 5, from High Specialty Regional Hospital of Oaxaca (HRAEO, internal dataset) and the National Cancer Institute (INCAN, external dataset), were analyzed. The morphology was analyzed using a circularity descriptor (к), and the texture was analyzed using the mean height/width ratio of the extrema descriptor (ρ). These results were compared with cancer/benign histopathology, which was binarily classified using ANNs. The best internal testing results were obtained with a one-hidden-layer ANN with 100 neurons, which successfully classified all high-density mammograms from patients with breast cancer in our internal dataset. External testing consistently yielded lower evaluation metrics across all ANN models, with reductions of at least half. The proposed morphology (к) and texture (ρ) descriptors show promise for detecting breast cancer in raw mammograms, with radiological findings, in a local context. However, their poor external performance highlights the need for substantial further work before this approach can be deemed suitable for broader diagnostic applications

Featured Image

Photo by Sandip Kalal on Unsplash

Photo by Sandip Kalal on Unsplash

Why is it important?

Research has recently been conducted using an Oaxaca population from Mexico. We found four rare breast cancers from HRAEO histopathology: one inflammatory breast cancer with an international frequency of < 5% of all tumors, one apocrine carcinoma, one malignant phyllodes tumor with a global frequency of < 1% of all tumors, and one secretory carcinoma with a worldwide frequency of < 0.1% of all tumors. We proposed analyzing whole-breast regions in mammograms because breast cancer alters the morphology and textural patterns of breast tissue. A large number of features can be extracted from radiological images. Still, only morphology (к) and texture (ρ) descriptors were employed as a carefully selected set of features, since this approach may yield a more robust and clinically interpretable model. As a result, we conducted a proof-of-concept study to assess whether these descriptors could effectively classify mammograms. This machine learning technique employs traditional computer vision analysis and leverages an ANN, using a smaller database than deep learning techniques.

Perspectives