Short on time?

Get essay writing help

Automated Melanoma Recognition in Dermoscopy Images Via Very Deep Residual Networks: Analytical Essay

  • Words: 1709
  • |
  • Pages: 4
  • This essay sample was donated by a student to help the academic community. Papers provided by EduBirdie writers usually outdo students' samples.


Automated melanoma recognition in using image processing technique from the available dermoscopic images in deep learning is difficult task because of the contrast and variation of melanoma in skin. It is mainly a non-invasive method so that it cannot contact with skin more forcefully. To overcome the disadvantages this paper proposes a method using very deep convolutional neural networks (CNNs).on comparing with older version of methods, in that lower-level features of CNN is employed. For more accurate classification in this method, we are using FCRN and CNN with the effective training limited data. Initially, Performance of Segmentation is done using residual networks using a image from the dataset followed by Classification by neural networks to check the abnormalities in skin. In this kind of classification technique, the network has more specified features from the segmented portion alone. The proposed technique is mainly evaluated on datasets and experimented with results that would show the performance in histogram and stages of melanoma which includes PSNR ratio.


Melanoma is actually a skin cancer, its initial stage is pigmenting on skin cells known as melanocytes. It is curable at the initial stage. It can be cured easily at starting stages. The major effect of this is from ultraviolet radiation with lower levels of pigmentation of skin. Lesions are mostly smaller in size and initially look like small dot.

Using the trained set of data, images get recognized and undergo FCRN which segments gradually and then using that enhanced image it will be introduced into classification to obtain the accurate results. Doing Pre-processing Techniques comprises of filtration, Enhancement. Performing Segmentation using FCRN and classification based on CNN. Showing any abnormalities while classification can be displayed at different levels.

Literature survey:

In [1] Deep Learning is implemented equipped with GPU (Graphics Processing Unit). Non-dermoscopic images which are taken from digital cameras which is referred to as clinical images. Clinical Images also known as non-dermoscopic images from which dermatologists recognize melanoma lesions and detect skin cancer. The non-dermoscopic images may contain noise effects and other factors such as hair so that they are eliminated by the method know as Pre-processing. Further, the Enhanced image or the output from the pre-processing unit is fed into CNN (Convolutional Neural Network). The advantage of this is diagnosis of skin cancer through Computer and it is a type of automated analysis. But it requires sophisticated algorithms. This is only for professionals and cannot help non-specialists. He concluded that cameras played a tool role in differentiating images. Here a complexity based on clinical images. As a result it increases accuracy and differentiation. In case of training a small data set is being used.

In [2] He says that idea on melanoma detection segments by 19-layer. CNN trained one end to other ends. In this medical segmentation, task is accurate. Advantageous of its effectiveness and efficiency. It is prone to suffer from variances. The results he arrived at Several effective training methods was implemented to overcome the challenges. Our approach methods when evaluating on an open challenge database of Skin Lesion Analysis Towards Melanoma Detection. This is vigorously arising to imaging with processing(both stages).

In [3]melanoma in automation recognition is part of deep leaning and which is derivation of CNN. Types of characteristics is being used for classifying. The main outcomes demonstrate high classification of very accurate compared with other datasets. The dataset contains hundreds of images. He concluded as accurate in classification is the best when compared with others.. These results are very promising and derived from CNN have high potential in melanoma recognition. In future work, we will creating a new model for Mat Conv Net.

In [4] RGB colour space is transformed is that whatever the colour information present in the images can be used for the purpose to differentiating the actual with affected. The most quantitative analysis on 250 images showed algorithm is efficient. in such a way the image intensity of each pixel is calculated through a weighted combination of the three RGB channels. This is effective for its speed function and flexible segmentation. One of the major disadvantageous is deformable models are normally semi-automatic the initial parameter values is used. This paper’s outcome is in this algorithm, it remains to be same unchanged. Further scopes for accurate shape and robust.

In [5] an easy and effective pre-processing for classification of melanoma by properties. the major part is aligned axis of tumour in simultaneous direction and (CNN) classified into the validation in 5-fold. The main advantage is to attain better in comparing with targeted property. Complexity is they require a minimum amount of trained datasets.

In [6] Chiranjeev Sagarl and Lalit gave their idea about melanoma as it is located in lesions which are detected. The melanoma detection can be done using colour spaces. With the help of colour spaces most important and necessary information regarding lesions which is embedded colour channel makes way for segmentation of a digital image. It is performed in clinical images. The image from mobile cameras does not have proper clarity The method used here is Colour channel-based Segmentation. Early diagnosis can cure melanoma in skin lesions and it is more accurate. Dermoscopic images not easy to get. Sophisticated algorithms are provided which is available only for professionals. This concludes as very simpler manner and easier manner on histogram. colour channels have been analyzed in different ways of the skin. The result is 94% accurate.

Proposed methodology:

Trained data are the images which act as an input for the process. The trained datasets are from ISIC where dermoscopic images is available. As image processing technique accepts trained images so that it can communicate easily with computer in software. The data sets are nothing but a skin lesion pigmented which may or may not be a melanoma. Median filter is used for filtration. It is used to eliminate noises present in the biomedical images. This is very simple and easy for implementing the smoother images. Also intensity is varied between one of the pixels. This is main in reduction of noise such as hair etc. so median filter normally reduces salt and pepper known as black and white noises that increase pixel and enhance the contrast and also increase intensity. Here, Adaptive Histogram Equalisation is helpful in increasing contrast of the images.

tage of the proposed FCRN is that it can make pixel-wise predictions, which is of valuable significance for skin lesion segmentation task.

Save your time!
We can take care of your essay
  • Proper editing and formatting
  • Free revision, title page, and bibliography
  • Flexible prices and money-back guarantee
Place Order

Further FCRN improves the segmentation and gives the portion of affected lesion. The trained datasets results in effective way more accurately in the CNN. Convolutional Neural network is used for classification which gives good and accurate result compared to conventional methods. After the segmentation it preceeds over the classification process, here the images undergoes up sampling and down sampling process through deeper networks and gives the accurate result through histogram representation.

Results and discussion:

The main aim of the project is to recognize melanoma using Deep Learning Algorithm. Here Deep Learning Algorithm refers as a subclass of machine learning. Therefore FCRN (Fully Convolution Residual Networks). This is used in segmentation under deep layers for data representations. This experiment is carried on Intel Windows 7 MATLAB R2014b software. The Hierarchy starts from Pre-Processing methods, Segmentation followed by Classification of an input image. The dataset that was downloaded from the website:

This website provides ISIC melanoma skin images of analysing the tools for automation of segmentation and examination for the skin lesions with the aim of accurate melanoma detection from dermascopic images.

A particular image (.tif,.jpg,.png) is taken as an input for process from the trained dataset for testing. Since the dimensions of the image is not convinient so that it will adjust into the Command Window so it is resize the image in the workspace. In order to reduce to pixel in given image and the input image (RGB 24 bit) is converted into salt and pepper noise that is grey coloured image is (8 bit). It doesn’t have clarity in its appearance therefore, greyscale image is filtered using a Median filter which removes noise that results some visibility of pixels through our naked eye. With the filtered image, Contrast Enhancement is performed using Adaptive Histogram Equalisation that would enhance the interior portion of skin lesion and increased the contrast of the image.

Segmentation is performed using FCRN (Fully Convolutional Residual Networks) technique. The iterations goes upto 100 and enhanced for final segmentation. This segmented image is then classified using CNN (Convolutional Neural Network) based on abnormalities at different levels that is plotted in Histogram as final results.

CNN Output:


Here, peak to signal ratio is measured from which it attains a peak value which is denoted by point.


This contains training, validation, test and best levels at which the image is classified. Up to 4 epochs or duty cycles, the classification takes place. These epochs or duty cycles can be changed based on the requirements.

Command window output

It will denote the stages as early stage, Moderate and Severe based on the type of Melanoma images. PSNR value appears in the command window.

Further scope:

Further scopes of this include integrating probability distributions for complexity areas. Also, it enhances the differentiate the performance and improves this method on future applications.


This paper proposes an image processing method for automated recognition which is based on very deep Convolution neural networks in order to proceed the recognition which consists of two steps: segmentation and classification. By framing two steps that without interaction of manually operated. On Comparing our method with other convolutions the CNNs can execute the data with higher differentiation through segmentation and classification. Further proceeding with a segmenting using FCRN gives an accurate result. Deep CNNs accompanied with trained sets which help to solve complexity in analysis of medical problems, with help of limited training data.. This will be more useful in easy detection. It is more useful in medical field for identifying skin cancer types at initial stage.


  1. Melanoma Detection by Analysis of Clinical Images Using Convolutional Neural Network -E. Nasr-Esfahani
  2. Automatic Skin Lesion Segmentation Using Deep Fully Convolutional Networks with Jaccard Distance-Yading Yuan
  3. Combining Deep Learning and Hand-Crafted Features for Skin Lesion Classification- Yading Yuan*, Ming Chao, and Yeh-Chi Lo
  4. A Novel Approach to Segment Skin Lesions in Dermoscopic Images Based on a Deformable Model -Zhen Ma and João Manuel R. S. Tavares
  5. Simple and Effective Pre-processing for Automated Melanoma Discrimination based on Cytological Findings -Takuya Yoshida

Make sure you submit a unique essay

Our writers will provide you with an essay sample written from scratch: any topic, any deadline, any instructions.

Cite this Page

Automated Melanoma Recognition in Dermoscopy Images Via Very Deep Residual Networks: Analytical Essay. (2022, September 27). Edubirdie. Retrieved December 1, 2022, from
“Automated Melanoma Recognition in Dermoscopy Images Via Very Deep Residual Networks: Analytical Essay.” Edubirdie, 27 Sept. 2022,
Automated Melanoma Recognition in Dermoscopy Images Via Very Deep Residual Networks: Analytical Essay. [online]. Available at: <> [Accessed 1 Dec. 2022].
Automated Melanoma Recognition in Dermoscopy Images Via Very Deep Residual Networks: Analytical Essay [Internet]. Edubirdie. 2022 Sept 27 [cited 2022 Dec 1]. Available from:
Join 100k satisfied students
  • Get original paper written according to your instructions
  • Save time for what matters most
hire writer

Fair Use Policy

EduBirdie considers academic integrity to be the essential part of the learning process and does not support any violation of the academic standards. Should you have any questions regarding our Fair Use Policy or become aware of any violations, please do not hesitate to contact us via

Check it out!
search Stuck on your essay?

We are here 24/7 to write your paper in as fast as 3 hours.