Segmentation and classification of consumer-grade and dermoscopic skin cancer images using hybrid textural analysis

Abstract

We present a skin lesion diagnosis system that segments the lesion and classifies it as melanoma or nonmelanoma. The proposed system is capable to deal with skin lesion images acquired by standard consumer-grade cameras and dermascopes. In order to suppress the image artifacts and enhance the lesion area, we propose an illumination correction strategy which consists of filtering in frequency and spatial domains. We introduce a hybrid model for lesion segmentation, which forms texture segments of the illumination corrected image using a factorization technique. Then based on the texture distinctiveness of the corrected and the texture segmented images, the saliency maps are computed, which are combined to decide lesion texture segments. In order to classify the segmented lesion, we propose a multimodal feature set composed of texture-, shape-, and color-based features. Classification performance of the multimodal features is evaluated using support vector machine, decision trees, and Mahalanobis distance classifiers. We evaluate the performance of the proposed system qualitatively and quantitatively. For the consumer-grade camera skin images dataset and ISIC 2017 dermascopic images dataset, the average segmentation accuracies are 98.4% and 95.4%, respectively; the classification accuracies are 98.06% and 93.95%, respectively.

Keywords: consumer-grade skin images; dermascopic images; illumination correction; multimodal feature set; saliency maps; skin cancer; texture segments.

Fig. 1

(a), (b) Examples of dermascopic images taken from ISIC grand challenge 2017 dataset. (c), (d) Examples of consumer-grade skin images.

Fig. 2

Visual results of the proposed illumination correction technique: (a) input image, (b) homomorphic filtering result on $L$ component in CIELab space, (c) homomorphic filtering output in RGB space, and (d) bilateral filtered image.

Fig. 3

Visual results of hybrid segmentation model for a dermascopic image from ISIC 2017 dataset: (a) input illumination corrected image, (b) texture segments image, (c) saliency map of illumination corrected image, (d) saliency map of texture segments image, (e) final mask obtained by proposed model, (f) GT binary mask, and (g) segmented lesion by proposed model.

Fig. 4

Visual results of hybrid segmentation model for a consumer-graded camera skin lesion photograph: (a) input illumination corrected image, (b) corresponding texture segments image, (c) saliency map of the input corrected image, (d) saliency map of texture segments image, (e) the final saliency map, (f) binary mask obtained based on postprocessing of (c), (g) binary mask obtained based on the postprocessing of (e), (h) GT binary mask, and (i) the masked lesion using binary mask of input corrected image of (a).

Fig. 5

Tonal variations in YIQ space: (a) skin lesion in YIQ space, (b)–(d) intensity surface plots of the Y, I, and Q channels, respectively. The vertical axis of (b)–(d) represents the intensity, whereas the axes labeled as $x$ and $y$ indicate horizontal and vertical dimensions of the segmented image.

Fig. 6

Scree graph for the proposed multimodal features extracted for the ISIC 2017 and the DermIS–DermQuest datasets.

Fig. 7

The visual results for the consumer-grade skin images of the proposed hybrid illumination correction technique are given in (f). For the sake of comparison, column (a) shows the original images. The visual results obtained by existing illumination correction techniques: (b) retinex, (c) morphological, (d) automated prescreening by Cavalcanti and Scharcanski, and (e) MSIM.

Fig. 8

Visual comparison of the segmentation results: (a) input images, (b) segmentation obtained using the binary masks of the proposed technique, (c) segmentation obtained using the available GT binary masks, and (d) visual differences between the binary masks used in segmentation of (b) and (c).

Fig. 9

Visual comparison of the segmentation results: (a) input images, (b) illumination corrected images, (c) segmentation results of proposed technique using illumination corrected images, (d) segmentation results of proposed technique without illumination correction, and (e) segmentation results using GT.

Fig. 10

(a)–(g) Visual comparison of the segmentation results of the proposed and state-of-the-art segmentation techniques.

Fig. 11

Receiver operating curves for the DermIS–DermQuest dataset results given in Table 11: (a) for SVM and (b) for decision tree classifier.

Fig. 12

Receiver operating curves for ISIC 2017 dataset: (a) for SVM and (b) for decision tree.