Figure 1. Percentage of study dermatologists who chose to biopsy melanomas before and after the receipt of the multispectral digital skin lesion analysis (MSDSLA) information. The overall biopsy choice of 69% before receipt of the MSDSLA information improved to 94% after information receipt.
Figure 2. Effect of multispectral digital skin lesion analysis (MSDSLA) information on biopsy decision for melanoma. Choice to biopsy all 5 of the melanomas rose from 13% of the dermatologists before MSDSLA information receipt to 70% after information receipt.
Rigel DS, Roy M, Yoo J, Cockerell CJ, Robinson JK, White R. Impact of Guidance From a Computer-Aided Multispectral Digital Skin Lesion Analysis Device on Decision to Biopsy Lesions Clinically Suggestive of Melanoma. Arch Dermatol. 2012;148(4):541–543. doi:10.1001/archdermatol.2011.3388
A major challenge faced daily by clinical dermatologists is to determine which pigmented lesions are appropriate for biopsy. The present study was designed to determine the effect of guidance provided by a multispectral digital skin lesion analysis (MSDSLA) device (MelaFind; MELA Sciences Inc)1 on dermatologists' decision to biopsy a pigmented lesion and the impact of the information provided by the device on the associated melanoma biopsy sensitivity and specificity. MelaFind uses light from visible to near-infrared wavelengths to image up to 2.5 mm beneath the skin and analyzes images from subbands of these wavelengths to provide information about the lesion's level of structural disorder. The device provides an output of “positive” or “negative” as an additional piece of data that can be integrated into the biopsy decision.
A total of 179 practicing dermatologists (median duration of practice, 11-15 years) attending an educational conference participated in an interactive melanoma session. Participants were asked to evaluate 24 pigmented lesions (5 melanomas and 19 other pigmented lesions) that had been analyzed as part of a prior study using a MSDSLA system.2 To make the experience more clinically realistic, the lesions were grouped from 4 composite patients, each having 6 lesions, with matching historic and clinical characteristics. Patient histories were presented, and then distant and close-up clinical and dermoscopic images of each of the lesions were viewed.
Each dermatologist responded yes or no on an electronic keypad to the following question: “Would you biopsy this lesion?” Then, the MSDSLA system information was provided, and the participant responded to this question: “Would you now biopsy this lesion?” Individual responses before and after MSDSLA information were compared to determine the effect of the MSDSLA information on the biopsy decision. The study was deemed exempt by the institutional review board of New York University.
For 179 dermatologists, the MSDSLA information improved the average biopsy sensitivity for the 5 melanomas from 69% prior to receiving the MSDSLA information to 94% after receiving the information (P < .001) (Figure 1). Biopsy specificity declined from 54% before to 40% after MSDSLA information receipt (P < .001). Biopsy rates of lesions that were MSDSLA negative fell from 43% before to 25% after MSDSLA information receipt (P < .01). Of the 4 lesions that were not evaluable by the MSDSLA system, biopsy rates went from 37% before to 42% after the dermatologists learned that no MSDSLA information would be available (P = .16, showing neither a positive nor negative effect when the system provided no additional information).
Integration of the MSDSLA data into the biopsy decision process also led to a more uniform decision by the dermatologists. The multirater κ statistic for interobserver agreement improved from 0.32 before to 0.45 (fair to moderate) after receipt of the additional information provided from the MSDSLA system.
The changes in biopsy decisions made as a result of integrating the MSDSLA device increased the overall biopsy sensitivity with a concomitant lesser decrease in overall biopsy specificity that was related in part to the fact that the device had identified as positive for biopsy some lesions that were not melanomas. However, for MSDSLA-negative lesions that were also histologically negative, the specificity increased. The overall impact of the intervention resulted in the number of potentially unnecessary biopsies being reduced by 17%.
Melanoma biopsy sensitivity for dermatologists has been demonstrated to consistently be in the range of 70% to 80%.3- 6 Therefore, there is an opportunity for computer-aided devices to improve selection accuracy for biopsying pigmented lesions. A prior prospective study found the measured sensitivity of MelaFind to be 98.4%.2
The present study demonstrated that having MSDSLA information to integrate into the biopsy decision led to a significant improvement in biopsying melanomas (69% to 94%). As would be expected, there was also a concomitant decrease in biopsy specificity after information receipt, from 54% to 40%, indicating that a biopsy recommendation from the MSDSLA system led to an increase in the number of biopsies of nonmalignant lesions. However, in the clinical setting, the risk of making a type II error (not biopsying a lesion that is a malignant melanoma) has far more severe consequences than a type I error (biopsying a nonmalignant lesion), so that in the case of melanoma diagnosis, improvement in sensitivity was more clinically relevant than a smaller decrease in specificity. When information was lacking from the MSDSLA because 4 lesions were not evaluable for technical reasons, there was no impact on the decision to biopsy (37% before vs 42% after the dermatologists learned that no MSDSLA information would be available), nor was it detrimental.
Finally, this study also demonstrates that when the additional MSDSLA information was provided, although it was incorporated into the decision process, it was not universally followed. After receiving the MSDLSA information, 70% of the dermatologists chose to biopsy all 5 of the melanomas (Figure 2), and 25% of the lesions reported as negative by the device were still biopsied. These facts demonstrated that the information provided by this type of diagnostic device for pigmented lesions was integrated into all of the other factors used in making a biopsy decision but was not “blindly” followed.
Until there is a cure for melanoma, early detection of disease remains an imperative goal to improve patient outcomes. While certainly there is no substitute for clinical judgment in assessing which lesions require further investigation, the present study suggests that having the additional information provided by a MSDSLA device can significantly impact the accuracy of the decisions made by dermatologists when determining whether a pigmented lesion should be biopsied.
Correspondence: Dr Rigel, 35 E 35th St, Ste 208, New York, NY 10016 (firstname.lastname@example.org).
Published Online: February 20, 2012. doi:10.1001/archdermatol.2011.3388
Author Contributions: Dr Rigel and Ms Roy had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Rigel, Roy, and Cockerell. Acquisition of data: Rigel, Roy, and White. Analysis and interpretation of data: Rigel, Yoo, Robinson, and White. Drafting of the manuscript: Rigel, Yoo, and Cockerell. Critical revision of the manuscript for important intellectual content: Rigel, Roy, Robinson, and White. Statistical analysis: Rigel, Yoo, Cockerell, and White. Administrative, technical, and material support: Roy, Robinson, and White.
Financial Disclosure: Drs Rigel and Cockerell serve as consultants for MELA Sciences Inc; Ms Roy is employed by MELA Sciences Inc.
Disclaimer: Dr Robinson is the Editor of the Archives of Dermatology, but she was not involved in the editorial evaluation or editorial decision to accept this work for publication.