Page 82 - AIH-1-3

Artificial Intelligence in Health Rotational thermography for breast cancer screening

explored. Results indicate the system’s promise as an the results and should be considered when interpreting
effective diagnostic tool for early detection of high-risk the findings. One potential source of bias arises from the
individuals. The system’s non-invasive and non-contact sample population, which may not fully represent the
nature makes it well-suited for population screening, general population due to demographic variations and
despite challenges with ambient temperature adjustments. differences in breast cancer prevalence, thereby potentially

After considering the experts’ concerns, we conducted limiting the study’s generalizability.
a thorough comparative analysis, including using the Confounding variables such as variations in breast
system with biopsy and USG. This analysis was carried out density, tissue composition, and patient positioning during
with great attention to detail. This analysis addresses the imaging could have affected the accuracy and consistency
expert’s and concerned doctors’ request for a comparison of the system. Although the study attempted to control
with established diagnostic methods. The system exhibited for these factors, they may have introduced some degree
a sensitivity comparable to that of biopsy (90.2% vs. 88.6%) of variability in the results. External factors such as
and USG (90.2% vs. 89.7%). Furthermore, the system equipment quality and maintenance, technician expertise,
displayed competitive specificity, with respective values of and interpretation differences among medical professionals
82.8%, 84.6%, and 82.4% for biopsy, USG, and the system. may have also impacted the study’s outcomes. Moreover,
This comprehensive analysis not only addresses reviewers’ relying on USG and biopsy reports to cross-validate the
concerns but also underscores the potential utility of the system’s performance introduces potential dependencies
system as a valuable diagnostic tool. on the accuracy and reliability of these other diagnostic
modalities.
5. Conclusion
The study progressed iteratively, with each phase’s
IR imaging plays a crucial role in various medical findings and feedback shaping the design and objectives
applications, emphasizing IR image acquisition of the subsequent phase. This approach allowed us to
techniques. This paper reported different types of IR image refine methods and address challenges progressively. By
acquisition systems based on trials in a hospital setup enhancing techniques and analyses based on results-
and conclusively identified the superior one. Key features driven objectives, each phase naturally evolved, focusing
of the proposed imaging system include a touchless and on enhancing the efficiency and accuracy of our breast
painless IR camera-based system for maximum patient cancer screening imaging system. In terms of sample size,
comfort; gantry rotation for acquiring multiple breast we acknowledge that the varying sizes across phases may
angles; dynamic IR image collection within a temperature- impact the overall consistency of the results. However,
controlled chamber; and a simplified user interface for the number of subjects available for each phase depended
data collection by technicians, IR image analysis experts, on live patient availability during the study period at the
and doctors. hospital. Practical constraints such as time and resource
The primary challenge in this study was managing limitations influenced sample sizes, despite efforts to
patients with varied health conditions. During the early maintain consistency. While we tried to work with
stages of development, patients had to wait a long time consistent sample sizes, external factors such as patient
to reach a stable room temperature. However, as the availability and medical considerations posed challenges.
development of the novel data acquisition technique However, our phased approach enabled us to optimize
progressed, the evolved system became more user-friendly methods and techniques, yielding improved results in
and efficient regarding imaging quality. The next challenge each subsequent phase. Our study focused on real-world
addressed was maintaining a constant ambient temperature application and practical implementation, requiring
during data collection, which was the most difficult task. It flexibility in our approach.
was overcome by implementing a temperature-controllable In addition, the challenges faced during data collection,
enclosure. Here, two ambient temperatures have been such as maintaining a constant ambient temperature and
taken. The higher temperature was 25°C, and the lower managing patients with diverse health conditions, affected
temperature was 23°C. The mean temperature adjustment the precision and consistency of the imaging process.
for each subject is 1°C. While we addressed these challenges throughout the study,
Finally, IR image analysis software was developed, residual variability may have affected the results. Overall,
incorporating machine learning algorithms that produced while the study presents promising findings, we carefully
excellent results. These findings were cross-validated using considered its limitations and potential sources of bias
USG and biopsy reports. However, several limitations when evaluating the system’s effectiveness and applicability
were identified during this study that may have influenced in broader clinical settings. Future research should address

Volume 1 Issue 3 (2024) 76 doi: 10.36922/aih.3312