Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Computer-aided diagnosis for cutaneous melanoma is a rapidly emerging but promising field of research that has the potential to help physicians to improve diagnostic accuracy for melanoma.
The combination of rising incidence of melanomas and the need to improve diagnostic accuracy have spurred the research in this field.
Sequential steps are involved in creating a computer-assisted analysis system. These include image acquisition, lesions segmentation, feature extraction and lesion classification.
Current computer-assisted analysis systems use differing algorithms and specifications.
Early detection is critical in lowering mortality from melanoma. As incidence and mortality continue to escalate, and because definitive therapy for advanced melanoma remains elusive, clinicians strive to evolve in their ability to diagnose melanoma at earlier, surgically resectable stages. New technologies, ranging from simple techniques such as photography to more complicated computer-based ones may improve clinical acumen. Ultimately, the role of computer-based diagnostic devices should be to increase the clinician's sensitivity by enhancing the ability to detect lesions and remove them while simultaneously maintaining a high specificity and avoiding the removal of benign lesions. While there is still much work to be done, there are many promising devices which use differing computer-augmented techniques that may serve to enhance the traditional dermatologist visual examination ( Table 37.1 ).
Technology | Sensitivity | Specificity | Advantages | Disadvantages |
---|---|---|---|---|
MoleMax | N/A | N/A | Two-camera system; no oil immersion required; transparent overlay for follow-up; total body photography | No computer diagnostic analysis |
MelaFind | 95–100% | 70–85% | Multispectral sequence of images created in < 3 s; handheld scanner | – |
Spectrophotometric intracutaneous analysis | 83–96% | 80–87% | Diagnosis of lesions up to 2 mm; visualizes skin structure, vascular composition and reticular pigment networks; handheld scanner | – |
SolarScan | 91% | 68% | Empirical database for comparison; session and image-level calibration; recorded on graphical map of body | Requires oil immersion |
Confocal scanning laser microscopy (CSLM) | 98% | 98% | Histopathological evaluation at bedside with similar criteria; longer wavelengths can measure up to papillary dermis; fiber-optic imaging allows for flexible handheld devices | Poor resolution of chromatin patterns, nuclear contours and nucleoli; can only assess to depth of 300 μm; melanomas without in-situ component will likely escape detection |
Optical coherence tomography (OCT) | N/A | N/A | High-resolution cross-sectional images resembling histopathological section of skin; higher resolution than ultrasound and greater detection depth than CSLM | Photons are scattered more than once, which can lead to image artifacts; ointment or glycerol may be needed to reduce scattering and increase detection depth; visualization of architectural changes and not single cells |
Ultrasound technology | 99% | 99% | Cost-effective; information about inflammatory processes of skin in relationship to nerves and vessels | Tumor thickness may be overestimated due to underlying inflammatory infiltrate; melanoma metastasis cannot be separated from that of another tumor |
Electrical bio-impedance | 92–100% | 67–80% | Complete examination lasts 7 minutes | Electrical impedance properties of human skin vary significantly with the body location, age, gender, and season |
The use of TBDI to document and monitor patients with many potentially dysplastic nevi has been well established. Here, images capture the morphology of existing nevi with standardized lighting and body poses allowing the clinician to ‘zoom in’ for closer evaluation of a concerning lesion. Serial evaluation with photos allows the clinician to better determine relative change in existing nevi as well as to identify new nevi.
Several computer-based imaging systems provide high-definition photographs that can be analyzed allowing for easier tracking of atypical nevi and the appearance of new pigmented lesions. With MoleMapCD (DigitalDerm Inc, Columbia, SC), patients are sent to medical photographers for 36 high-resolution images documenting the patient's entire skin surface using standardized lighting and poses ( Fig. 37.1 ). These images are then stored on a CD. One is given to the physician and one provided to the patients for self-skin examinations. Images on the CD have embedded software which allows the clinician to enlarge and look closer at concerning nevi.
MIRROR DermaGraphix (Canfield Scientific, Fairfield, NJ) has software that enables the linking of close-up or dermoscopy images to specific points on a patient's regional body sector photos. The MIRROR DermaGraphix software is designed for image acquisition performed by, or under the supervision of, each practitioner. In this case, individual, regional, or total body digital images are obtained and downloaded into the software for storage and viewing. A CD or total body photography book can be generated and provided to the patient to enhance self-skin examination.
A side-by-side comparison between total body or regional photographs and the patient skin lesions are completed by the medical provider in the office and/or the patients during self-skin examinations. This method is time-consuming and subjective to individual physician experience and style. In an attempt to automatize TBDI examination and detect changes objectively, a collaborative project between the University of Arizona and Raytheon Missile System was initiated in 2005. To date, a proof of concept has been completed by integrating sophisticated registration and change detection algorithms used in the remote sensing field to standardized TBDI images. Consecutive images can thus be objectively compared for change ( Fig. 37.2 ). As envisioned, the proposed system will generate quantitative results in size and pigmentation that can be compared over time.
Dermoscopy is being increasingly adopted as a melanoma diagnostic technique ( Chapter 36 ). Several computer-based algorithms have recently been developed to help with dermoscopy interpretation; however, none has been shown to be superior to physician-based methods. Blum et al. created an algorithm using 64 analytic parameters evaluating 837 digital images of benign lesions and melanoma proving to have an accuracy comparable to that by dermoscopic experts. Others have seen similar results.
One caveat regarding the assessment of computer-based dermoscopy is the possibility of photo selection bias of the images used for creating and testing the algorithms. A recent study of three computer-program-driven diagnostic instruments found significant variability in the diagnostic accuracy of the instruments in the evaluation of suspect melanocytic lesions. While the three systems were able to accurately identify clinically obvious melanoma, they tended to incorrectly classify most seborrheic keratoses as potential malignant lesions.
Tools are also being developed to allow comparison of dermatoscopic images across time. The DermoGenius Ultra (LINOS AG, BIOCAM, Gottingen) takes and uses standardized images to quantify dermoscopic characteristics into a dermatological point score. These images and scores are compared over time to determine if significant change has occurred.
Become a Clinical Tree membership for Full access and enjoy Unlimited articles
If you are a member. Log in here