Vol.:(0123456789) 1 3
Journal of Ambient Intelligence and Humanized Computing
https://doi.org/10.1007/s12652-018-1051-5
ORIGINAL RESEARCH
Skin lesion segmentation and recognition using multichannel saliency
estimation and M-SVM on selected serially fused features
Tallha Akram
1
· Muhammad Attique Khan
2
· Muhammad Sharif
3
· Mussarat Yasmin
3
Received: 10 March 2018 / Accepted: 15 September 2018
© Springer-Verlag GmbH Germany, part of Springer Nature 2018
Abstract
The number of deaths caused by melanoma has increased remarkably in the last few years which are the carcinogenic type of
skin cancer. Lately, computer based methods are introduced which are intelligent enough to support dermatologist in initial
judgment of skin lesion. However, there still exists a gap for an optimal solution; therefore, machine learning community is
still considering it a great challenge. The primary objective of this article is to efciently detect and classify skin lesion with
the utilization of an improved segmentation and feature selection criteria. Presented contribution is threefold; First, ternary
color spaces are exploited to separate foreground from the background—utilizing multilevel approach of contrast stretching.
Second, a weighting criterion is designed which is able to select the best solution based on extended texture feature analysis,
related labels, boundary connections and central distance. Third, an improved feature extraction and dimensionality reduction
criteria is proposed which combines conventional as well as recent feature extraction techniques. The proposed method is
tested on PH2, ISBI 2016 and ISIC benchmark data sets and evaluated on the basis of multiple parameters including FPR,
sensitivity, specifcity, FNR and accuracy. From the statistics, it is quite clear that the proposed method outperforms numer-
ous existing techniques with considerable margin.
Keywords Skin cancer · Image enhancement · Image fusion · Feature extraction · Feature selection
1 Introduction
Melanoma is one of the deadliest types of cancers respon-
sible for massive number of deaths worldwide (Codella
et al. 2018; Oliveira et al. 2018). Although it accounts
for approximately 3–4% of all skin cancer cases, still it is
the primary source of death of around 75% people related
to skin cancers (Codella et al. 2018). In America alone,
178,560 melanoma cases are reported in 2018 including
91,270 invasive and 87,290 non invasive cases. Whereas
the estimated total deaths in 2018 are 9320 containing 3330
women and 5990 men (Siegel et al. 2018). In Australia, an
estimated number of skin cancer cases treated in 2018 are
14,320 including 8653 males and 5667 females. Estimated
number of deaths in 2018 due to skin cancer are 1905
including 1331 males and 574 females (Read et al. 2018).
Discriminating benign (a common nevi) and melanoma (a
lethal skin cancer) at their early stages is quite difcult as they
both exhibit similar features at early evolutional phase which
is one of the main problems for an expert dermatologist. The
primary reasons behind are morphological structural factors
of lesion like color spectrum, shades of color lesion skin gra-
dient, multiple lesions, streaks and grid patterns (Hendi and
Martinez 2011). Figure 1 shows skin lesion images taken from
two diferent data sets. The severity of pigmented skin cancer
and its resulting demise can be curtailed if it is detected at an
early step of inception. Melanoma is almost curable and sur-
vival rate increases if it is detected when the malignanttumor
thickness is less than a specifc threshold and at its very early
stage (Brunssen et al. 2017; Sng et al. 2009).
In clinical interpretation, the doctors utilize dermoscopy
for lesion detection. Dermoscopy is a non invasive method
used to study the skin lesion in dermatology. In clinical
examination, the most commonly visual inspection methods
are ABCD rule (Kawahara et al. 2018; Monisha et al. 2018),
* Mussarat Yasmin
mussaratabdullah@gmail.com
1
Department of Electrical Engineering, COMSATS
University Islamabad, Wah Campus, Wah Cantt, Pakistan
2
Department of Compute Science and Engineering, HITEC
Univesity, Museum Road, Taxila, Pakistan
3
Department of Computer Science, COMSATS University
Islamabad, Wah Campus, Wah Cantt, Pakistan