Discrimination of cancerous from benign pigmented skin lesions based on multispectral autofluorescence lifetime imaging dermoscopy and machine learning

Abstract

Significance: Accurate early diagnosis of malignant skin lesions is critical in providing adequate and timely treatment; unfortunately, initial clinical evaluation of similar-looking benign and malignant skin lesions can result in missed diagnosis of malignant lesions and unnecessary biopsy of benign ones.

Aim: To develop and validate a label-free and objective image-guided strategy for the clinical evaluation of suspicious pigmented skin lesions based on multispectral autofluorescence lifetime imaging (maFLIM) dermoscopy.

Approach: We tested the hypothesis that maFLIM-derived autofluorescence global features can be used in machine-learning (ML) models to discriminate malignant from benign pigmented skin lesions. Clinical widefield maFLIM dermoscopy imaging of 41 benign and 19 malignant pigmented skin lesions from 30 patients were acquired prior to tissue biopsy sampling. Three different pools of global image-level maFLIM features were extracted: multispectral intensity, time-domain biexponential, and frequency-domain phasor features. The classification potential of each feature pool to discriminate benign versus malignant pigmented skin lesions was evaluated by training quadratic discriminant analysis (QDA) classification models and applying a leave-one-patient-out cross-validation strategy.

Results: Classification performance estimates obtained after unbiased feature selection were as follows: 68% sensitivity and 80% specificity with the phasor feature pool, 84% sensitivity, and 71% specificity with the biexponential feature pool, and 84% sensitivity and 32% specificity with the intensity feature pool. Ensemble combinations of QDA models trained with phasor and biexponential features yielded sensitivity of 84% and specificity of 90%, outperforming all other models considered.

Conclusions: Simple classification ML models based on time-resolved (biexponential and phasor) autofluorescence global features extracted from maFLIM dermoscopy images have the potential to provide objective discrimination of malignant from benign pigmented lesions. ML-assisted maFLIM dermoscopy could potentially assist with the clinical evaluation of suspicious lesions and the identification of those patients benefiting the most from biopsy examination.

Keywords: autofluorescence; computer-aided diagnosis; feature selection; fluorescence lifetime imaging; machine learning; skin cancer.

Fig. 1

Summary of methodology showing maFLIM image acquisition, preprocessing, feature extraction, and classification. maFLIM, multispectral autofluorescence lifetime imaging.

Fig. 2

(a) Transformations in a single pixel multispectral maFLIM data during pixel-level preprocessing. (b) Example maFLIM image with $K$-means cluster mask and the two separated regions. The images map the total integrated intensity of the maFLIM signals at each pixel location. maFLIM, multispectral autofluorescence lifetime imaging.

Fig. 3

Transition of a sample maFLIM image to the corresponding 2D histogram distribution on the phasor plot. Figure also shows the transformation of pixels from both regions 1 and 2 on the maFLIM image into corresponding points on the phasor plot computed at an arbitrary frequency component.

Fig. 4

(a) 2D histogram phasor distributions from the pixels corresponding to the two regions in an maFLIM image. The distance between the distributions is indicated by “$d$.” (b) Phasor distribution scatter plots with bivariate Gaussian fits on regions 1 and 2. The covariance matrices ${\mathrm{\Sigma }}_1$ and ${\mathrm{\Sigma }}_2$ give a measure of spread of the two regions and $\theta$ represents the angle between their major axes. (c) Phasor distribution scatter plot showing the variances ${\sigma }_p^2$ and ${\sigma }_q^2$ along the major axes. The ratio of the variances indicates the symmetry of the distribution.

Fig. 5

Flow diagram showing (a) feature selection process using LOPO-CV along with SFS algorithm, and (b) detailed steps involved in the SFS algorithm. The number of features selected, $n_{\mathrm{SFS}}$, is varied from 1 to 7 for all feature pools and 1 to 6 for intensity feature pool. SFS, sequential forward search; LOPO-CV, leave-one-patient-out cross-validation; AUC,– area under the curve.

Fig. 6

(a) Schematic of classification of skin lesions. The posterior probabilities from the individual classifiers are combined in an ensemble fashion. (b) Weight optimization for the ensemble classifier. The optimum weight is selected from the ROC curve. LOPO-CV, leave-one-patient-out cross-validation; QDA, quadratic discriminant analysis; ROC, receiver operator characteristics.

Fig. 7

(a) Handheld maFLIM dermoscope imaging the forearm of a patient. (b) Clinical photograph of a melanoma lesion. (c) Time-domain maFLIM feature maps of a melanoma lesion. The columns show the feature maps corresponding to the three emission channels. First row shows the weight of the fast decay. Second row shows the fast lifetime maps, while the third row shows the slow lifetime maps. Average lifetime maps are shown in the fourth row. Fifth row shows the integrated intensity maps of each spectral emission channel, and the ratio of the intensities are shown in the sixth row. The last row shows the cluster mask generated for the lesion and the integrated intensities from all the channels for the clustered regions 1 and 2. The horizontal strip in the images is due to the presence of hair on the skin during imaging.

Fig. 8

Histogram of weights on one of the feature pools, when combined in an ensemble fashion for (a) phasor-intensity, (b) biexponential-intensity, (c) phasor-biexponential, and (d) phasor-biexponential-intensity feature pools.