Page 72 - AIH-1-3

Artificial Intelligence in Health Rotational thermography for breast cancer screening

Table 1. A summary table of earlier works and a comparison with our approach
Study Findings Pros Cons Uniqueness of our study
Etehad Tavakol High detection rates using bi-spectral High detection rates; Limited scope; may Our study expands on bi-spectral
et al. 7,10 invariant features and K-means effective segmentation lack versatility across invariant features with more advanced
clustering for segmentation. 7,10 technique. different conditions. machine learning algorithms.
Garduño-Ramón Non-invasive tool utilizing temperature Non-invasive; promising May require further Our study includes more precise
et al. 11 and texture features, yielding promising outcomes. optimization for temperature controls and advanced
results. 11 diverse cases. segmentation techniques.
Various K-means, fuzzy c-means, and EM Superior accuracy with May face challenges in Our study integrates advanced
segmentation algorithms were explored, with EM EM algorithm. real-time applications. techniques, including fuzzy c-means
techniques showing superior accuracy. 11,12 clustering, for better real-time
outcomes.
Venkataramani Semi-automated method using High sensitivity and Semi-automated Our study uses a fully automated
et al. 10,13,18-22 morphological filtering and specificity. methods may still approach with robotic arm movement
thresholding, achieving high sensitivity require human and data processing.
and specificity. 10,13,18-22 intervention.
Deep High accuracy in segmenting High accuracy in Deep learning models Our study applies machine learning
learning-based suspicious regions in breast segmentation. may require large in tandem with a novel data collection
approaches thermograms. 9,23,24 datasets for training. protocol for more comprehensive
results.
General challenges Standardization of temperature values; Promising results in Difficulty in Our study introduces a non-contact,
in IR imaging need for trained personnel; variability detecting abnormalities. establishing precise non-invasive approach with precise
in individual heat sources. 28-34 temperature temperature control for improved
thresholds. results.
Abbreviations: EM: Expectation-maximization; IR: Infrared.

specificity. Neural network (NN) parameters and pattern of the breast from a particular angle, as the camera
recognition tools assessed the system’s performance with movement and rotation caused the images to overlap. This
high accuracy rates across different phases. resulted in the inner quadrant of one breast’s IR image
being superimposed by the other breast’s outer quadrant.
2. Data and methods
Accordingly, a significant modification was made to
2.1. Data collection address this concern in PS3, as illustrated in Figure 1C.
In PS1, data collection was undertaken following the The patient was seated in a fixed position, and the camera
technique previously described in the literature. 14,35 moved in a semi-circular arc on an arm-based arrangement.
The subject was seated in front of the camera with both A tabletop mechanical arrangement was developed to
hands raised upward, as illustrated in Figure 1A. In PS2, ensure that only one breast was focused at the pivot point of
the subject was seated similarly to PS1 with two cameras the semi-circular arc. The second breast was isolated from
deployed at two corners for IR image acquisition. 14,30,36 the camera view by covering it with an IR-proof barrier.
Figure 1B illustrates the setup for the same. A tabletop setup of PS3 is shown in Figure 2.
Several modifications were carried out in the imaging Finally, rotational thermography was set up in the FS.
process based on the doctors’ guidance. One modification A camera rotated in a semi-circular arc and stopped at
was placing a camera in a fixed position with the patient different angles, with the patient seated in a fixed position.
seated on a rotating chair. In this case, images were acquired The arrangement is illustrated in Figure 3.
from various predetermined angles. The data collection It comprises an enclosed chamber with a breast-shaped
angles are 0°, 30°, 60°, 90°, 120°, 150°, and 180° from the grooved hole through the chamber wall. The patients’
initial position. Since the patient was rotating, focusing on breasts are positioned one at a time through this hole for IR
a specific breast was challenging. It also led to a shifting imaging. The distance from the camera to the subject is 1 m,
region of interest (ROI) in the images. This variation in the minimum focus distance of the IR camera calibrated
focal length led to unsatisfactory imaging results. by the manufacturer at a thermal laboratory. The ground
The subsequent logical adaptation was repositioning clearance of the system was 0.76 m. A total of 32 thermal
the camera while keeping the patient stationary. However, IR images were acquired for each subject, with 16 images
a significant challenge arose in obtaining a clear IR image acquired at a higher ambient temperature, for example,

Volume 1 Issue 3 (2024) 66 doi: 10.36922/aih.3312