Towards Automated Diagnosis of Basal Cell Carcinoma M. Larraona-Puy a , A.Ghita a , A. Zoladek a , W. Perkins b , S. Varma b , I.H. Leach c , A. A. Koloydenko d , H. Williams e and I. Notingher a a School of Physics and Astronomy, University of Nottingham, University Park, NG7 2RD, Nottingham, U.K b Dermatology Department, Nottingham University Hospital NHS Trust, QMC Campus, Derby Road, NG7 2UH, Nottingham, U.K. c Histopathology Department, Nottingham University Hospital NHS Trust, QMC Campus, Derby Road, NG7 2UH, Nottingham, U.K. d Mathematics Department, Royal Holloway, University of London, Egham, TW20 0EX, U.K. e Centre of Evidence-Based Dermatology, C Floor South Block, Nottingham University Hospital NHS Trust, QMC Campus, Derby Road, NG7 2UH , Nottingham, U.K. Basal Cell Carcinoma (BCC) is the most common malignancy in humans and constitutes about 8 out of 10 diagnosed skin cancers. For large, rare or recurrent BCC’s and for those growing into the areas where maximum healthy tissue conservation is required, Mohs micrographic surgery (MMS) is considered the most suitable treatment. The main limitation of MMS is the visual border discrimination of the tumor, which is currently performed by histopathology examination. 1 Histopathology diagnosis is a non-automated, subjective, invasive and time- consuming technique. As it has already been reported, Raman micro-spectroscopy (RMS) can be used as a powerful non-invasive objective alternative tool in discriminating BCC from its non- tumorous surroundings. Although for the last decade several studies have proved the capability of Raman spectroscopy to distinguish between BCC, epidermis and dermis by analyzing the differences in their Raman spectra, 2-4 the creation of Raman maps showing the accurate location and automated diagnosis of all these three classes within a skin scanned region is required. 5 In this preliminary study an approach to the problem is proposed using a supervised multivariate algorithm for imaging and diagnosis of BCC. Selected Raman bands responsible for the largest spectral differences between BCC and normal skin regions and linear discriminant analysis (LDA) were used to build the classification model. A total of 329 skin-tissue Raman-spectra from 20 patients allowed BCC discrimination from healthy tissue with 90±9% sensitivity and 85±9% specificity in a 70% to 30% split cross-validation algorithm. This multivariate model was then applied on unknown tissue sections from new patients to image tumor regions. Several images from in vitro sections both with the presence and absence of BCC have been mapped using our proposed algorithm. These maps have been compared with their corresponding hispathology examinated H&E images, achieving satisfactory