P55: A noninvasive screening tool for the detection of melanoma in a publicly available ISIC dataset using a single parameter in hue saturation values space

R.S. Prasad ¹ and V. Prasad²

¹DSITM College, Dr APJ Abdul Kalam Technical University, Ghaziabad, Uttar Pradesh, India and ²Department of Nuclear Medicine, Ulm University, Ulm, Germany

The mortality and quality of life of a patient with melanoma can be positively influenced by a noninvasive, highly sensitive screening tool. Apart from sensitivity, robustness and ease of implementation influence the success of a screening tool. The hypothesis of this spectral analyses-based study is that the colour intensity of skin lesions holds the key for detection of melanoma in skin lesions. With this background, a MATLAB-based novel diagnostic screening tool for the manual diagnosis of melanoma is proposed. Original images of skin lesions with a histopathologically confirmed diagnosis of melanoma and naevus were collected from a publicly available dataset (ISIC). The pixels of images were manually retrieved on careful visual observation. Thereafter, the red–green–blue (RGB) data of the pixels were converted into hue saturation values data. Scatter of saturation (S) data against pixel locations on images showed the unique feature that all malignant lesions were either confined within a threshold of an S value not greater than 0·40, an S value of > 0·60, or confined partly in both regions. A rule (named ‘RSP’) was then framed that if > 50% of S data were found to be confined within the two threshold regions, melanoma was confirmed. A total of 250 lesions were analysed of which 138 were melanoma and 112 were naevi. RSP correctly diagnosed all melanoma lesions [true positive (TP)], whereas 23 naevi were falsely characterized as melanoma [false positive (FP)] the remaining 89 were correctly diagnosed as naevi [true negative (TN)]. None of the melanoma lesions was missed by the RSP. Thus, the RSP achieved 100% sensitivity and a specificity [TN/(TN + FP)] of 79·5%. The average overall length of time from the start of data collection from the ISIC archive followed by data processing and finalization of the report for each image was only 25 min. The RSP-based identification method offers a unique possibility to detect melanoma with 100% sensitivity, thereby fulfilling the basic criteria of a screening tool. Further studies are warranted to assess this simple and easy-to-use screening test in an unbiased reader-blinded trial. If validated, its diagnostic performance can easily be transformed into an independent automatic melanoma identification algorithm as an adjunct to a dermatoscope.