top of page
ultrasound1.jpg

Cancer Sign Assessment from Ultrasound Images using Computer Visions

Methods for analyzing medical images utilizing both traditional machine learning and deep learning approaches.
2.png

01

Cancer Lesion ROI

Specialised radiologists usually mark cancer lesions in ultrasound images by a set of the region of interest (ROI) points (red dots in the image).

02

Border Irregularity Recognition

Medical professionals evaluate the lesion border irregularity as a crucial factor in diagnosing cancer. The border irregularity strongly indicates lesion malignancy. The border irregularity can be measured through Euclidean distances between border points and various reference points or shapes. The figures on the left depict two instances of lesions with irregular borders.

397_1.png
21_interpolatedPoints.png
image.png
388.png
397.png
65.png

03

Distances to Different Reference Shapes

Border irregularity is measured by evaluating a distance function calculated from boundary points (marked with red dots) to various fitted shapes like ellipses, convex hulls, and Gaussian curves (indicated by yellow curves). After normalization, the distances (depicted as white lines) are employed to construct a distance function.

04

Distance Function

The distances between border points are interpolated (dashed blue curve), which is then used for more in-depth analysis of irregularities. The minimum and maximum points of the distance curve, marked with red pluses, are used to visualize the locations of irregularities as indicated by red stripes.

FDindex: 1.0036
FDindex: 1.0614
FDindex: 1.1971

05

Method Inspired by Fractal Dimensions

The perimeter of the nodule border, represented by the green curve, is measured at various scales similar to Fractal Dimensions. These scales are determined by selecting varying numbers of border points (such as 20, 45, 90, etc.) for perimeter calculation resulting in several perimeter values. Subsequently, the perimeter values are divided by one perimeter value of the fitted ellipse, resulting in an index denoting irregularity (FDindex). This FDindex serves to characterize the intense of irregularity and can be employed for classification purposes.

bottom of page