Page 81 - AIH-1-3

Artificial Intelligence in Health Rotational thermography for breast cancer screening

Figure 9. Neural network classification results of the training and testing datasets for breast abnormality detection for the final study (88 subjects). The
results display model tuning and performance evaluation. The image was created using Matlab software.
Abbreviations: CE: Cross-entropy; %E: Percentage of correctly classified elements.

method was rigorously applied across all 88 subjects in the abnormalities, providing a different perspective on breast
FS phase. cancer screening. Although integrating these traditional
Focused on IR imaging, the study recorded the number measures could enhance the study, our concentration was
of pixels corresponding to each temperature cluster zone on advancing the field of IR imaging to contribute valuable
and used this information to quantify areas of interest. knowledge to breast cancer screening.
Given the 640 × 480-pixel IR images, each subject’s dataset The system’s repeatability was confirmed by imaging the
included seven zones across two ambient temperatures for same breast 5 times, demonstrating high consistency with
16 images (total number of image data = [71 × 2] + [10 × 4] + minimal variability. Results across trials were consistent,
[33 × 32] + [88 × 32] = 4054). as evidenced by acceptable statistical measures, including
The study’s primary focus was to explore innovative IR standard deviation and coefficient of variation, affirming
imaging techniques and related features for breast cancer the system’s accuracy and clinical viability for breast cancer
screening. We aimed to investigate temperature-based screening.
imaging and machine-learning algorithms as alternative Clinical implications and utility: the findings regarding
diagnostic methods. This approach allowed us to detect their clinical implications and the system’s utility in breast
thermal patterns and variations that could indicate potential cancer diagnosis and population screening have been

Volume 1 Issue 3 (2024) 75 doi: 10.36922/aih.3312