Can We Improve Breast Cancer Diagnosis by Using Ai-Based Algorithms in Low and Middle-Income Countries: The First Moroccan Experience on A Private Dataset

H. El Agouri *¹, M. Khmou ¹, Y. Mehdi ¹, M. Azizi ², H. El Attar ³, M. El Khannoussi ², S. EchCharif ¹, C. Mounjid ⁴, A. Souadka ⁵, B. El Khannoussi ¹

1. Pathology department, Oncology National Institute, Faculty of Medicine and Pharmacy, Mohammed V University, 10100 Rabat, Morocco.

2. Datapathology, 20000 Casablanca, Morocco.

3. Anatomic pathology laboratory Ennassr, 24000 El Jadida, Morocco

4. Pathology department, Oncology National Institute, Faculty of Sciences, Mohammed V University, 10100 Rabat, Morocco.

5. Surgical department, Oncology National Institute, Mohammed V University, 10100 Rabat, Morocco.

*Correspondence to: H. El Agouri. Pathology department, Oncology National Institute, Faculty of Medicine and Pharmacy, Mohammed V University, 10100 Rabat, Morocco.

Copyright

© 2023 H. El Agouri. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Received: 20 July 2023

      Published: 01 August 2023

Abstract

We are currently living in the middle of a digital revolution; pathological anatomy is no longer an exception, particularly in breast oncology. Indeed, the diagnosis and classification of breast cancer using Deep learning algorithms have attracted a lot of attention. Yet, there are many limitations towards their implementation in low and middle-income countries. In Morocco, we have succeeded through the continuity of our project, initiated in January 2020, to make our database larger and our model more efficient. In the current work, we worked on the extension of our dataset by collecting overall 1229 digital slides in order to improve classification accuracy. Our approach utilizes several deep neural network architectures and

gradient boosted trees classifier. We report high degrees of overall correct classification accuracy and sensitivity for detection of carcinoma cases. This approach would solve a global public health problem in areas with a shortage of pathologists.

According to the World Health Organization (WHO), breast cancer (BC) is considered a critical public health issue and a leading cause of cancer-related deaths among women worldwide, especially in low and middle-income countries (LMIC) (1). Additionally, pathologists are closely involved is establishing an accurate and early diagnosis of BC, which is essential for improving treatment efficacy and increasing survival rates.

In Morocco, both public and private laboratories are facing an increasing caseload, while the number of pathologists is decreasing for various reasons. These reasons include a higher prevalence of cancer associated with aging populations and the implementation of improved cancer screening programs. Add to this, a considerable number of pathologists are approaching retirement, and there is a limited availability of diagnostic pathology training for young doctors. This situation has often led to overwork, tiredness, and a higher risk of diagnostic errors, ultimately resulting in significant delays in cancer diagnosis. In this context, the use of artificial intelligence (AI) algorithms has gained increased attention in field of pathology. AI assisted diagnoses have the potential to be employed in daily pathology practice, offering benefits such as enhanced workflow efficiency, improved accuracy, and better reproducibility.

As a middle-income country, the current circumstances makes Morocco a unique target for the widespread implementation of digital pathology (DP) and adoption of AI-based algorithms in breast surgical pathology. These advancements can greatly assist in decision-making and encourage personalized treatment, which have become paramount in BC diagnosis and management. However, their development is mainly restricted by significant costs issues, inadequate infrastructure and lack of standardization. Moreover, the absence of national data as well as the limited training of medical personnel in many small pathology departments, due to the low rate of acceptance toward this change.

Currently, the digital revolution is transforming the practice of breast surgical pathology worldwide. The application of DP based on AI algorithms has primarily been assessed in high-income countries. However, its implementation in low-resource and rural settings, which face an exponential increase in cancer cases and a shortage of trained pathologists, remains sporadic (2). Nevertheless, in this digital era of globalization, the transition to the digitization of pathology services should not be limited to developed countries. The adoption of AI-based models is expected to be more efficient in low and middle-income countries (LMICs), offering multiple advantages to pathologists and patients. These advantages include improving diagnostic accuracy, obtaining second opinions, and guiding treatments, particularly in the post-COVID-19 era.

Despite the concerns and challenges, we were the first pathologists in North Africa to introduce AI-based tools in diagnostic pathology through our project, which was initiated in January 2020. As a result, we have conducted a groundbreaking study titled "Assessment of deep learning algorithms to predict histopathological diagnosis of breast cancer: first Moroccan prospective study on a private dataset" (3). In this study, we have successfully demonstrated the powerful classification capability of our algorithms in terms of BC histopathology images, utilizing our private and unique dataset.

Although our promising initial results, we were challenged by the limited size of our dataset compared to ones publicly available. To address this, we focused on expanding our dataset by collecting a total of 1229 digital slides. This expansion was crucial to enhance the accuracy of our classification. In our approach, we employed a hybrid pipeline that incorporated data augmentation techniques, as well as supervised and unsupervised machine learning algorithms. This methodology was previously described in our study (3).

The experimental results demonstrated high levels of overall correct classification accuracy (92%) and sensitivity (94%) in detecting carcinoma cases. These findings hold significant importance for diagnostic pathology workflow as they assist pathologists in making early and accurate BC diagnoses. Moreover, our results exhibited a strong discriminatory power, both in distinguishing between benign and malignant cases and classifying the three sub-categories. Overall, our results are highly encouraging and comparable to the current state-of-the-art applications of deep learning models in histological diagnosis of BC (4), (5).

In summary, it is undeniable that the future of pathology is digital, and the integration of DP and AI-based algorithms into diagnostic practice is no longer a luxury but a necessity, particularly in LMICs. In Morocco, the successful implementation of these technologies requires the full engagement of pathologists, oncologists, specialists, and data scientists. Furthermore, their deployment should be an integral part of a comprehensive national strategy, including the establishment of a national electronic healthcare system. This will enable pathologists to embrace AI, fully utilize its capabilities, and become pioneers in cancer management.

Reference

1. World Health Organization facts on breast cancer. https://www.who.  /cancer/prevention/diagnosis-screening/breast-cancer/en/.

2. Andrey Bychkov MD, PhD, at the 2022 annual meeting of United States and Canadian Academy of Pathology (USCAP)

3. El Agouri H, Azizi M, El Attar H, et al. Assessment of deep learning algorithms to predict histopathological diagnosis of breast cancer: first Moroccan prospective study on a private dataset. BMC Res Notes. 2022;15(1):66. Published 2022 Feb 19. doi:10.1186/s13104-022-05936-1

4. Peyret R, Pozin N, Sockeel S, et al. Multicenter automatic detection of invasive carcinoma on breast whole slide images. PLOS Digit Health. 2023;2(2):e0000091. Published 2023 Feb 28. doi:10.1371/journal.pdig.0000091
5. Khairi SSM, Bakar MAA, Alias MA, et al. Deep Learning on Histopathology Images for Breast Cancer Classification: A Bibliometric Analysis. Healthcare (Basel). 2021;10(1):10. Published 2021 Dec 22. doi:10.3390/healthcare10010010.