How computer vision works in medicine: let’s look at an example
Ultrasound is the primary method for diagnosing breast cancer, and the speed of this diagnosis, together with accuracy, has a direct impact on cure rates and mortality. Analyzing ultrasound images is one of the most challenging tasks, so special machine learning techniques are now being applied to biomedical images to achieve high classification rates – ones those would replace humans and could “look” at medical images with the eyes of an experienced doctor.
The healthcare industry, for example, uses high-tech cameras and sensors to visualize and extract information to implement this idea. But this is only a small part of computer vision technology. A more complete range of its capabilities can be assessed by the greed-based deep feature generation based image classification model.
A group of scientists working with clinics in China, Turkey and Saudi Arabia shared the data of their research, which showed a successful model for diagnosing ultrasound images.
What does the work of an effective model look like?
The framework involved 16 pretrained models as feature generators. In the first step, the input image used is divided into rows and columns and the generators are applied to each row and column. The model can calculate the error value of each deep feature generator, then it selects the best three feature vectors and considers them as the final ones.
At the feature selection stage, iterative neighborhood component analysis (INCA) selects the optimal 980 features. Finally, these features are classified using a deep neural network (DNN).
The scientists concluded that the developed grid-based image classification model achieved an accuracy of 97.18% in detecting ultrasound images for three classes of breast tissue changes, namely malignant, benign and normal. This accuracy gives the method the potential to extract information from medical images that may not be visible to the human eye.
It is only logical that such results support the trend of increased investment in computer vision technology development. Facts and Factors market research report predicts the market size to grow up to $2.4 billion by 2026.