Deslandes's study involved a comprehensive review of literature and the evaluation of transvaginal ultrasounds (TVUS) by six professionals, who assessed 150 images of the uterus and ovaries.
The results of this evaluation revealed only poor to moderate agreement among the observers regarding the quality of the images, indicating a pressing need for further refinement of the scoring system.
Although AI has the potential to assess image quality more objectively than humans, the success of these systems hinges on human labeling, which may introduce noisy data due to subjective assessments.
As part of this initiative, PhD student Alison Deslandes is working on developing a quality scoring system specifically for gynecological images utilized in AI diagnostic algorithms.
This subjectivity in interpreting ultrasound image quality presents challenges for consistent scoring, which could ultimately affect the training of AI algorithms.
Research from the University of Adelaide has unveiled the promising role of artificial intelligence (AI) in diagnosing women's health conditions, particularly endometriosis.
Ultimately, the effectiveness of AI tools in ultrasound diagnosis is contingent upon high-quality data, as poor-quality images can significantly impede the performance of deep learning systems.
