Early detection of breast cancer through screening is the most effective method for reducing cancer related morbidity and mortality. Despite its benefits, mammography has limited sensitivity especially in dense breasts. Among the recent advances aiming to combat the limitations of mammography is the use of AI .
In 1990s, CAD (computer aided detection) was introduced to increase radiologists accuracy in detection and diagnosis of breast cancer. But no true clinical benefit was achieved as a result. Eversince, there has been an ongoing research to develop more advanced technology, such as deep learning. Deep learning models are not limited by the human understanding of what a breast cancer looks like, but rather teach themselves what to look for after being exposed to multiple examples of normal and pathological images .
In the current work, AI achieved higher sensitivity than mammography in detecting the different types of malignant breast lesions and the commonest carcinoma was IDC (68%), Fig. 1.
The sensitivity of AI was 96.6%, and false negative rate 3.4%, while mammographic sensitivity was 87.3% and false negative rate 12.7%. AI performed better than mammography in detecting different histopathological types of breast malignancy namely DCIS, IDC and ILC with sensitivity (100%, 96.7%, 96.6%) vs (88.9%, 89%, 82.2%) respectively. While in other rare types of breast malignancy (micropapillary carcinoma, borderline phyllodes tumor and tubular cribriform carcinoma), both AI and mammography showed the same sensitivity 80%.
The current results were consistent with multi-institutional studies such as the one performed by Kim et al.  in South Korea, the USA, and the UK. They reported that AI standalone sensitivity in the three validation datasets was 91%. Another retrospective study based on the screening programe in Western Australia reported that AI based systems outperfomed human radiologists with 14.2% increase in senitivity .
Close results were attained by Ribli et al.  where the AI emloyed system achieved AUC = 0.95, (95 percentile interval: 0.91 to 0.98, estimated from 10,000 bootstrap samples). Meanwhile, Watanabe et al.  reported that with the aid of an AI system, the cancer detection rate among radiologists increased from an average of 51% to 62%.
Our findings are also agreeing with smaller scale studies [11,12,13,14,15,16]. Rodriguez-Ruiz et al.  in 2019, performed a study on 240 mammograms (100 cancers, 40 leading to false-positive recalls, 100 normal) reported that sensitivity increased with AI support to 86% vs 83% with mammography alone. Other study by Pacilè et al.  in 2020, which was carried out on 240 participants, also reported that average sensitivity of radiologists was increased by 0.033 when using AI support (P = 0.021). Sasaki et al.  also suggested that AI can increase the sensitivity of human readers from 89 to 96%.
Although several studies discussed the role of AI in cancer detection among screening mammograms, few studies addressed its significance in detecting different types of cancer and their different mammographic appearances. Our study showed that AI was more sensitive to detect cancers with suspicious mass 95.2% vs 75%, suspicious calcifications 100% vs 86.5% and asymmetry and distortion 100% vs 84.6%, than mammography.
This study results went with Kim et al.  who reported AI sensitivity in detecting soft tissue lesions (mass, asymmetry and distortion) to be 89.8% vs 71.6% for reading mammograms by unassissted radiologists. Also in their study the AI showed better sensitivity than radiologists in detecting microcalcifications (87.6% vs 71.4%).
Another study also suggested that in malignant cases, AI performed better than mammography in both mass detection and calcifications. (84.1–86.1% vs 77.5–77.9%) . Furthermore, in Conant et al. study , sensitivity of AI in cases with only calcifications was 100% and was 88% in cases with soft-tissue densities with or without calcifications.
The AI system employed in the current study showed 100% senitivity in detection of DCIS. This could be explained through the pathological nature of DCIS where calcification is a prominent feature . Our study showed ability of AI to detect pathological calcification at a sensitivity rate of 100%.
In mammographic imaging, ILC doesn’t always present as a mass. It may have vague appearances in the form of architectural distortion, asymmetry, or even breast size discrepencies, making it one of the most missed pathological types of cancer on mammography (up to 30% of missed cancers). Mammographic sensitivity in detecting ILC ranges from 86 to 100% in fatty breasts down to 45–68% in extremely dense breasts . It was found in the current study that AI can overcome the limitations of mammography in cases of ILC, Fig. 2. It shows better sensitivity than mammography in detecting that type of carcinoma (96.7% vs 89%).
The applied color hue provided by the AI scanning delineated the full extent of the disease not just targeted the site of the abnormality on the mammogram . Such criteria helped to determine the multifocal or multicentric distribution of the included carcinomas, Figs. 2 and 3.
Regarding the histopathological types of breast cancer, this work was consistent with a study performed by Lang et al. . In agreement with our results, AI showed high sensitivity in detecting both IDC, ILC, and DCIS. The AI in their study missed three tubular carcinoma, two IDC and one ILC. Our results reflected more or less similar findings, as in the current work the AI algothrium missed three cases of IDC, one case of ILC, and one case of borderline phyllodes carcinoma.
However, the AI algorithm sometimes underestimate breast lesions that lack the morphologic features of malignancy on the mammogram. As a result, no demarcation by the color hue was noted at the heatmap images and the given abnormality scoring for the breast showed accordingly a “Low” scoring, Fig. 4. To get the best performance of AI in correlation with mammogram; complementary ultrasound is mandatory to decrease the incidence of false negative results in case of diagnostic mammogram or screening mammogram with high breast density .
Lang et al.  and our study both showed 100% sensitivity in detection of DCIS, which highlights AI’s advantage in detection of pathological microcalcifications.
This study has some limitations. It is limited by the relatively small sample size. We did not study the interaction of a human interpreter with AI algorithm results and how AI will affect radiologists’ final assessment. The used AI algorithm does not take into account clinical factors such as symptoms or family history, which may limit comprehensive analysis.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.