AI over Thermal Images for Medical Applications: Thermal Imaging History and Manual Interpretation Challenges

A Quick History of Thermal Imaging for Medical Applications

The use of body heat as an indicator of disease dates back to ancient times and was a common practice in various cultures. One of the earliest recorded uses of body heat as a diagnostic tool can be traced back to the ancient Greeks, particularly the physician Hippocrates (c. 460-370 BCE). Hippocrates, often referred to as the "Father of Medicine," believed in the importance of observing and recording a patient's symptoms, including fever or elevated body temperature, as part of the diagnostic process. 

Hippocrates and his followers recognized that certain diseases could be associated with changes in body temperature, and they believed that fever was a sign of the body's efforts to fight off illness. While they did not have the sophisticated tools that we have today, their observations laid the foundation for the understanding of body heat and temperature as important indicators of disease.

Till 19th century, it was only feasible to obtain point temperatures, i.e temperature at a particular spot and it was not possible to measure the temperature profile of an area. However, the mid 20th century, thermal imaging of an area was actively researched during World War II for the purposes of  military applications. The first practical thermal imaging systems, often using lead sulfide detectors, were first developed in 1950s. However, these early systems were large, costly and cumbersome.

After that, cooled thermal cameras came into existence in 1960s. These cameras required cryogenic cooling, often using liquid nitrogen, to operate effectively. Though they were characterized by their high sensitivity and were used in military, industrial, and scientific applications, use for medical imaging was still a challenge. 

It was only in 1990s, the currently used uncooled thermal cameras came into existence. These cameras eliminated the need for cryogenic cooling and offered advantages such as reduced size, cost, and complexity. In 2000s, we saw further improvements in camera size, cost, and performance. This is when researchers have again relooked it usage for medical application, as it became affordable and portable.

Manual Interpretation Challenges

Now, you might be wondering why thermal imaging isn't more widely used in clinical practice despite its numerous advantages. One significant factor contributing to this is the complexity involved in manually interpretation of thermal images. Manual interpretation  demands a high level of expertise, is subjective, and can be mentally taxing. Let's delve deeper into the challenges associated with this.

(i) Temperature Limits and Color Palette Visualization

In the last blog, we explored how thermal imaging cameras generate temperature array and color palettes are used to transform this data into thermal images. However, the generation of thermal images are highly dependent on the choice of color palette temperature limits Tmin and Tmax. Let us see the below images, both of which were obtained using the same temperature array and same rainbow color palette but with differing temperature limits Tmin and Tmax.

The image on the left has temperature limits (25°C,  36°C), while the image on the right has temperature limits (28°C, 36°C). If you examine region near the arrow marks, you will easily recognize abnormal activity on the right image, while you might perceive left image as normal. However, both are obtained from same temperature matrix. Upon further investigation with histopathology, it is found out that the area shown by arrow mark indeed coincide with breast malignancy. This is one major reason why it requires high expertise to perform visual interpretation with naked eye.

(ii) Normal/non-disease related Elevated Thermal Patterns 

One of the critical challenges in the interpretation of thermal images is the occurrence of elevated thermal patterns that may not necessarily be indicative of underlying diseases. It's essential to recognize that a myriad of factors can influence thermal patterns on an image, and not all of them are associated with pathological conditions. Several non-disease-related factors can also contribute to these elevated heat pattern:

• The human body naturally exhibits variations in temperature, often related to factors such as circadian rhythms, hormonal fluctuations, and localized increases in blood flow. These normal variations can appear as elevated thermal patterns on thermal images, making it vital to distinguish them from disease-related patterns.
• Certain benign conditions, such as inflammation, musculoskeletal injuries, or skin disorders, can lead to localized increases in temperature. For instance, an area of the body with recent physical trauma or a localized skin infection may exhibit elevated heat patterns without the presence of malignancy or pathology.
• External factors, such as the clothing worn by the individual or exposure to heat sources, can influence thermal patterns. Clothing can insulate the skin and affect heat dissipation, leading to localized differences in temperature that are not necessarily linked to disease
• Emotional stress, variations in environmental temperature, and even exercise can cause transient changes in thermal patterns. These fluctuations are typically not indicative of an underlying medical condition but can contribute to the complexity of thermal image interpretation.

(iii)  Loss in Temperature Precision with Color Images

The transformation of the temperature values into color representations might lead to a loss of temperature precision during the process of visual interpretation. This loss of precision happens due to the finite number of colors available in the chosen color palette.

For example, let's assume that the selected color palette encompasses only 64 color shades. If the body temperature range under examination spans 8°C, applying this palette to represent temperature variations leads to a loss of granularity in the conversion. Each shade of color now signifies a temperature increment of approximately 0.125°C (8°C/ 64 ). This, however, is notably coarser than the intrinsic thermal sensitivity of the camera, which may actually be as fine as 0.05°C.

Further, it's important to note that even if there are 64 shades available in the color palette, the human eye has its limitations when it comes to differentiating between these colors. Only a subset of these shades can be effectively distinguished with the naked eye. This limitation introduces an additional challenge, further eroding temperature precision during the visual interpretation process.

Moreover, a typical thermal image comprises hundreds of thousands of individual pixels, each representing a discrete temperature value. For medical professionals, such as physicians or thermographers, this can translate into a significant cognitive burden. The process of visually inspecting and analyzing each pixel across an entire thermal image can be a taxing endeavor. It requires a high level of attention to detail and can be prone to errors, both due to the limitations of human perception and the sheer volume of data to be processed.

(iii) Lack of Imaging Standardization

Standardization plays a pivotal role in guaranteeing the precision and validity of the images captured. The process of capturing thermal images may appear deceptively straightforward, leading to a wide array of imaging protocols being attempted by individuals.  This lack of uniformity has given rise to variations in image quality, consistency, and interpretation, making it challenging for medical professionals to rely on thermal imaging as a diagnostic tool. Further, it's imperative to understand that the lack of standardization in thermal imaging can significantly hinder its reliability and effectiveness in clinical practice. 

It is only recently that the thermography societies such as American Academy of Thermology are establishing and disseminating standardized protocols for thermal image capture across different anatomical regions. These protocols provide clear guidelines on aspects like camera settings, patient preparation, environmental conditions, and the specific techniques required for capturing thermal images effectively. By instituting standardized procedures, the medical community can benefit from more consistent and reliable thermal imaging data. 

AI to Rescue

The challenges inherent in manual interpretation have indeed led to a dichotomy in the results and findings related to the use of thermal imaging for various medical applications. Some studies and research endeavors demonstrate remarkably positive outcomes, highlighting the diagnostic potential of thermal imaging, while others reveal instances of poor performance and inconsistent results in the hands of human interpreters. This variance in findings underscores the intricacies and subjectivity involved in manual thermography.

This is precisely where the advent of Artificial Intelligence (AI) comes to the forefront as a powerful and promising solution. AI, with its computational prowess, has the capability to bridge the gap between unstructured temperature data and structured, quantifiable information. AI not only overcomes the challenges of manual interpretation but also enhances the accuracy, consistency, image standardization and efficiency of thermal imaging as a diagnostic tool. As a result, it holds the promise of expanding the applications of thermal imaging in various medical domains, offering new opportunities for early detection, disease management, and improved patient care.

Skepticism Surrounding Medical Thermography

If you are a researcher who is new to this innovative technology, you may encounter a fair share of skepticism and criticism. There is no need to be deterred by these doubts and concerns expressed by medical practitioners. The simplicity in the use of medical thermography and the bold claims about its benefits  prompted regulatory bodies to establish strict regulations back in the late 1960s, which led to a decline in its utilization. However, it's crucial to acknowledge the significant advances that have occurred both in imaging camera technology, as we previously discussed, and in the development of artificial intelligence (AI) algorithms. These advancements point towards a promising future for thermal imaging in clinical applications.

In fact, we're witnessing a burgeoning wave of startups championing the combination of thermal imaging and AI for various medical applications. This convergence of technologies is poised to usher in a new era in healthcare, where thermal imaging, backed by sophisticated AI algorithms, will contribute to earlier disease detection, improved patient care, and enhanced diagnostic capabilities. The landscape is evolving rapidly, and it's an exciting time for researchers and practitioners alike to explore the vast potential of this integrated approach in the medical field.

Upcoming Blogs

In the coming blogs, we delve into specific clinical use cases of thermal imaging.

Do post your questions in the comment section!