Detecting skin cancer using hyperspectral images

Hyperspectral imaging has the potential to better detect skin cancer to improve patient survival rates.

Skin cancer is one of the most common types of cancer and in some parts of the world it is becoming increasingly common. Most skin cancers are not life-threatening, but one particular type called malignant melanoma is known to be quite deadly. If malignant melanoma isn’t detected early, it can spread below the surface of the skin to the lymph nodes, bones, lungs, and brain.

Malignant melanoma is sometimes referred to as a “silent killer” because the spread of cancer cells occurs out of sight and symptoms may not appear until it has become metastatic. Metastatic melanoma has a 5-year survival rate of about 10%. However, if malignant melanoma is detected early, the 5-year survival rate increases to around 95%. In other words, detecting melanoma as early as possible is crucial to saving lives.

Correctly identifying malignant melanoma in vivo at an early stage is difficult for anyone who is not an expert in the field of dermatology. Recently, a new approach for the detection and classification of skin cancer has been explored. Instead of conventional digital images using measurements of the red, green and blue regions of the light spectrum, some research groups have explored the use of hyperspectral images to address the challenge of skin cancer.

Hyperspectral images are created using contiguous spectral sampling across a range of the light spectrum, often within and beyond the visible range. Researchers from UiT, Arctic University of Tromsø and University of Las Palmas de Gran Canaria recently explored the state of the art in the application of hyperspectral imaging for the detection of melanoma in a manuscript published in WIRE Computer Statistics. Their review suggests that the infrared region of the light spectrum is able to capture increased levels of hemoglobin and the ultraviolet region is able to capture increased water retention in cells. These two events are known to be biological indicators associated with cancer.

Rich information like this is not readily available to a dermatologist when using image-based diagnostics with conventional digital images captured using a so-called dermoscopic camera system. These specialized camera systems allow a dermatologist to capture high resolution digital images of a skin lesion or mole, usually in conjunction with a glass contact surface used to flatten the skin area. The dermoscopic images are then used as part of the diagnostic procedure.

Since the images are captured using conventional digital camera sensors, the information is limited to the spectrum of human-visible light and the surface of the skin lesion or mole. However, the information in a hyperspectral image also comes from regions outside of the human-visible light spectrum, and some of the measured light also comes from below the surface of the skin. The authors suggest that physicians’ access to more information provided by hyperspectral images, in conjunction with specialized computer-aided diagnostic systems, should improve their ability to detect skin cancer at an early stage.

Moreover, it should be possible to develop artificial intelligence systems capable of detecting and classifying skin cancers using hyperspectral images. Two of the significant obstacles are the lack of standardized equipment and image acquisition procedures. First, there is currently a wide range of hyperspectral imaging systems available, albeit extremely few, that have both been designed and approved for dermatological applications. Second, different systems operate in different regions of the light spectrum, varying in spectral and spatial resolution, and having different approaches to image acquisition. These differences complicate the collection of image datasets, the development of standard algorithms, and the comparison of published results.

Recent studies on using hyperspectral images to catch skin cancer as early as possible are promising. However, many questions remain open and several key challenges need to be addressed accordingly in order to make substantial progress towards significant improvements in current manual and semi-automated skin cancer diagnostic procedures.

Meeting these challenges is crucial because detecting and correctly classifying skin cancer as early as possible is vital to saving human lives. There are also substantial socio-economic costs associated with misclassifying a benign skin lesion or mole, as all suspected skin cancers are surgically removed and diagnosed by a pathologist. This process places a burden on the medical system and significantly affects its profitability. Finally, the staging and surgical removal of suspected skin cancer can be traumatic for the patient, both physically and mentally; therefore, it is essential to reduce misclassification rates.

Written by: Thomas Haugland Johansen, Kajsa Møllersen, Samuel Ortega, Himar Fabelo, Gustavo M. Callico and Fred Godtliebsen

Reference: TH Johansen, et. al. “Recent Advances in Hyperspectral Imaging for Melanoma Detection.” WIRE Computer Statistics (2020). DOI: 10.1002/wics.1465

Previous scientists analyze the final Cassini images | Research & Technology | January 2020
Next Breakthrough could lead to cameras capable of capturing images using ambient radio waves