-Aged men generally have a hypertrophied prostate. This hypertrophy can evolve in cancer or can just be a benign hyperplasia. Some recent studies revealed that a linear hyperpigmentation of skin called Linea nigra (LN), can appear in men developing prostate cancer or benign prostatic hyperplasia. LN's prevalence is higher in the case of prostate cancer. It has been established that when a LN appears on someone's suprapubic region and has a length greater than ten centimeters, it indicates a metastasized or metastasizing tumor. Based on this, we developed in a previous work a method of LN segmentation. From segmented images, we can calculate descriptive values such as length, width, area, texture and color. These values are LN characteristics that can be used by specialists to make diagnosis. In this paper, we propose an implementation of those methods in an Android mobile application. The aim is to connect patients to their specialists for distant monitoring purpose based on images taken with their mobile phones. The specific architecture of the platform also helps in providing more precision in detection thanks to a deep learning program.
INTRODUCTION
Usually found in pregnant women, the linea nigra (LN) is a
linear hyperpigmentation of skin that appears between navel
and suprapubic region. It sometimes appears in men of a
certain age and non-pregnant women. Generally, LN is found
on the abdomen of about 75% of women in pregnancy [
Medical images are one of the most widely used tools to
better understand both normal and abnormal processes that
affect health. In particular, image processing algorithms can
provide quantification, accuracy, reliability and repeatability of
measurements and analyzes by delegating tasks to computer.
Because of that, development of computer aided diagnostic
systems (CADS) is a research field of interest [
Recent advances in computing and Artificial Intelligence (AI) have had a huge impact on interpretation of medical images. However, since obtained results cannot be fully taken into account, it is essential to have final conclusions from doctors. It is for all these reasons and also in order to ensure a fast, cost-effective and remote consultation for everyone that we propose in this paper an online platform whose aim is to detect and to analyze pigmented disorders, while taking into account the expertise of distant specialists.
II.
Using mathematics and technological advances, scientists
created different applications to equip dermatologists. These
applications also help to raise awareness and bring medical
prowess closer to population. Among them, we can enumerate:
• A special skin cancer screening application developed by
SkinVision that can be downloaded to smartphone [
SkinVision’s application is used by in many european countries
Fig 1. System architecture
• Maryam Sadeghi developed Molescope, an application
that, using a smart phone connected to a mini-microscope,
allows to take very precise images of spots and moles in order
to detect possible malignant melanomas. [
• A team from University of Houston, USA, developed the
DermoScreen application, an application capable of preventing
skin cancer. By photographing a suspect lesion or mole with
the phone, the software analyzes the risks whether it is cancer,
with a success rate of 85%. [
III.
Applications conception mostly for mobiles have become for some years accessible to the community; that is possible to the presence and availability of numerous modeling and programming tools. However, in the frame of medical applications pointed, it is generally professionals we meet. The conceptual phase stands for to take into account every features and mathematical treatments to do on the images for LN evolution detection. The database modeling is edited with UML. The development phase of platform must follow MCV for ease the application maintenance.
People involved in using the platform are:
The system of LN detection application integrates in its functioning the mobile application associated with web interface for computers used by everybody as he is a doctor or patient (figure 1). The network cloud is made up by images processing and treatment algorithms implemented for LN evaluation.
A mobile application running on smartphone and using the network connectivity for rural populations mostly need to be soft in its use. It means that processes and technologies used to develop and deploy application must not be smartphone chip gluttonous.
Access to internet is generally difficult in rural and remote places. At the first side, the coverage of terrestrial mobile networks is not enough because the mobile networks operators judge useless to densify with base stations or allocate a great bandwidth for a land where mobile traffic is not important. At the other side, the bandwidth and data rate put available are inappropriated for increasing rural needs. In front of this fact, mobile API mustn’t be greedy in chip capacities and data.
Fig 2: Rural scenario of application
The mobile platform useful for remote people must integrate the basic functions like:
Fig 3: Basic mobile features
Sick people (patients) can sign up (register) on the platform. They can register with a generalist ID associated or a dermatologist associated. The first doctor associated to one patient must validate the enrolment before the continuation (the patient will have access to health features).
Medical personnel must sign up with mandatory informations useful to track their service states; The doctor treating a patient can add in need case till two specialists;
Tchat OTT (mini OTT) module has two modes: mode reserved (discussion group for one treated patient doctors) and mode open (discussion group between doctors college and a patient).
Patients can send LN images to the online platform from a mobile phone; As soon as an image is sent by a patient, his doctor is notified. From then on, doctor can connect to the platform and launch image analysis process; At the end of an analysis process, found characteristics can be used by specialists to make diagnosis. Sick patient receives advices and possibly prescriptions or an appointment for a thorough auscultation.
For technical choices, we bet on python technologies. Firstly, Python is multiplatform. Secondly, Python is one of best scripting languages with supported libraries in Image processing domain. The image processing module must be at the basis entirely in Python (opencv, numpy, matplotlib...). Thus, the details relative to mobile application using Python technologies are:
- Front-end: Kivy is the Python mobile API selected. Using Kivy, mobile application done can be distributed on PlayStore. The application can use also Android services (SMS, camera, mail and notifications system); and have access to most of the normal java API.
- Back-end: SL4A (Scripting Layer for Android). SL4A makes possible scripting languages to Android by using Python for complex processes.
- Web: The web interface accessible is guaranteed by Django utilization web version of application in order to allow to medical personnel to have flexibility in their technical works
The image processing module has been developed in
python programming language based on LN’s segmentation
algorithm proposed in [
Once LN images has been segmented, some characteristic values can be calculated. They are: length, width, ratio, area, and histogram of oriented gradients.
They are determined from two furthest points considering respectively X axis and Y axis.
It is the ratio of width to length, gives an indication of the form of the pigmented disorder. Indeed, a ratio very close to zero indicates a very sharp LN. A value of ratio close to 0.5 indicates a LN more or less wide and / or not very long. A ratio close to 1 indicates a regular shape: circle or square.
It corresponds to the total number of pixels that belong to segmented region.
This descriptor calculates local histograms of the gradient orientation on a dense grid (evenly distributed areas) of the images. First proposed in [] the method has proven effectiveness for the detection of people.
RESULTS Fig4: Some LN segmentation results that Doctors can view
Fig5: Graph showing a segmented LN dimensions
Two segmentation results are shown in figure 4. Since it is important from clinical point of view to circumscribe the whole NL, we can conclude that our method is good. Considering the segmented LN of figure 5, the following characteristic values
The development of a technology combining a large amount of information on skin disorders and the analysis of data calculated from images can increase the information available for Dermatologists. This motivated our research on the development of a platform to serve as a diagnostic aid system. The developed platform is suitable for elderly people whose mobility possibilities become reduced with advanced age. It is also very useful for people living in rural areas.
In future work, we should focus on automatic detection of probable or non-probable metastasis of a prostate tumor based on identified characteristics combined with results of biomedical analyzes and clinical symptoms.