A new study has revealed that artificial intelligence (AI) can facilitate radiologists in better detection of breast cancer from mammography images. The Lancet Digital Health has published these findings online on February 06, 2020.
This study, carried out by Lunit Inc. and Korean academic hospitals, has increased the worth of AI in diagnosing breast tumors. Lunit is an artificial intelligence company that works for the development of AI-based solutions for oncology and radiology.
Five institutions across the UK, the USA, and South Korea have provided the data of more than 170,000 mammogram examinations, including the breast images of Caucasian and Asian females. Among this dataset, the researchers have also included the data of more than 36,000 independent cases that were identified with cancer using biopsy.
The dataset used for this study was almost seven times greater compared to that of previously conducted studies on this matter, said Hyo-Eun Kim. The study’s first author, Hyo-Eun Kim, is currently working as Chief Product Officer at Lunit Inc. to develop AI-based medical image analysis solutions.
In this study, the research team has compared the efficiency of radiologists in diagnosing breast cancers, before and after the use of artificial intelligence. The comparison showed an improvement in the radiologists’ performance after using AI.
According to the study findings, using AI alone in breast cancer detection correctly identified 88.8 percent of the cases (sensitivity). Whereas, the figure was 75.3 percent in the case of radiologists. One of the significant findings was that AI-assisted radiologists showed an increase in the sensitivity value from 75.3 to 84.8 percent.
ALSO READ: Longer Commute Time May Increase the Exposure to Carcinogens in Car Seats
Additionally, it was also found that in contrast to the radiologists, artificial intelligence (AI) had better sensitivity in diagnosing tumors with mass (78 percent vs 90 percent) and asymmetry or distortion (50 percent vs 90 percent).
Compared to radiologists alone, AI was also more efficient in diagnosing early-stage invasive cancers such as T1 cancers. The radiologists were able to detect 74 percent of both T1 and node-negative cancers. On the contrary, AI succeeded in detecting 87 percent of node-negative cancers and 91 percent of T1 cancers.
Breast density also plays a role in cancer detection using mammograms. A greater tissue density can act as an obstacle during cancer diagnosis, as it may mask the cancers in mammograms. This situation can be mostly observed in mammograms from the Asian population.
The results of the study revealed that breast density didn’t have much effect on the performance of AI for cancer detection. While, in contrast, the diagnostic performance of radiologists was much affected by this factor, showing a sensitivity of 79.2% for fatty breasts and 73.8% for the dense ones.
But on assistance by AI, the sensitivity of radiologists for diagnosing cancers in dense breasts was improved by 11 percent. Reducing the number of false-negative cases (missed ones) is one of the major concerns while detecting malignant tumors from mammography pictures.
To avoid these false-negative cases, the radiologists may incline towards increased recalls that may result in a large number of needless biopsies. Long-Term experience is required by radiologists for accurately interpreting breast images and the study reveals that the use of AI can identify tumors at an early stage and facilitate in detecting breast cancers with fewer recalls.