Michael Zuckerman-Visual system inspire algorithm for breast cancer risk assistant in terminal image

Electrical Engineering Systems Seminar

Speaker: Michael Zuckerman

M.Sc. student under the supervision of Dr. Hedva Spitzer

Sunday, 4^th February 2024, at 15:00

Room 011, Kitot Building, Faculty of Engineering

Visual system inspire algorithm for breast cancer risk assistant in terminal image

Abstract

In the last decade, breast cancer has become one of the leading cancer types worldwide. Every year, it causes the deaths of 685,000 individuals in the world. Early detection is crucial for increasing the chances of survival, and thermography is one tool physicians use to evaluate the risk of the disease. It is a noninvasive tool for measuring temperature distribution over the skin using infrared (IR) wavelets. 

Many previous studies tried to enhance thermal image through manly via denoising and histogram equalization. In recent years, several studies have attempted to improve the evaluation methods for distinguishing between sick and healthy patients.  

Our study aimed to improve the image appearance of the area in the image, such as areolar regions, nipples, and blood vessels (BV) that might give an indicator for breast cancer.

We propose a new algorithm that aims to enhance the above diagnostic areas through a unique algorithm for enhancing the structure of the image by lateral facilitation visual system mechanism and preprocess of companding HDR image based on Adaptive Contrast Companding (ACC) and a component that computes the region of interest (ROI) of the breast.   

The algorithm showed a 16% increase in the contrast of the blood vessel and areolar. It has been demonstrated on images from the DMA database through contrast-to-noise ratio (CNR) in comparison to the original BIR image.

We employed the deep learning evaluation method ResNet50 to validate our algorithm's effectiveness. This evaluation categorizes images into "sick" and "healthy" classes. The algorithm succeeded in obtaining better scores for the “sick” and the “healthy” cases (97%) in comparison to the ability to distinguish the diagnostic in the original image (86%). Our proposal algorithm yielded better accuracy and precision than the previous studies, at least for those that relied only on frontal direction breast examination.

השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בדף הנוכחות שיועבר באולם במהלך הסמינר