The assessment of computational intelligence technique for the diagnosis of acute lymphoblastic leukemia

Abstract

AimThis research project’s primary goal is to assess how accurate a computer vision-based system is at diagnosing acute lymphoblastic leukemia (ALL).MethodsIn the present study, artificial intelligence technology was used to diagnose ALL. Three deep machine learning models, which include VGG16, ResNet18, and ResNet34 with image augmentation techniques, were employed for the diagnosis of ALL. The digital images were acquired from the Giemsa-stained smears of anonymized cases of ALL. The study included 515 digitized images of ALL and 600 digital images from the normal blood smears of the anonymized cases. The obtained digital images were subdivided into three sets, which included a training set (60% of total images), a validation set (20% of total images), and a test set (20% of total images).ResultsThe present study was conducted on 1115 digital images which included 515 of ALL and 600 of normal blood cells. Among the three models of deep machine learning, the ResNet18 and ResNet34 models obtained 100% accuracy for categories of acute lymphoblastic leukemia and normal blood cells, while the VGG16 model obtained an accuracy of 99%.ConclusionThe application of artificial intelligence for the histological diagnosis of acute lymphoblastic leukemia revealed excellent results. In terms of accuracy and F1-score, the ResNet18 and ResNet34 models surpassed the VGG16 model. These models have the potential to be employed as an adjunct technique for the histological diagnosis of ALL, which would be helpful for better patient care.

Keywords

Introduction

The recent developments in digital technology have produced astounding results in a variety of areas, including healthcare. The advancement in digital technology has revealed a healthy impact on the upgradtion of patient care. Artificial intelligence makes a substantial contribution to the successful outcomes of digital technology. The computational technology is enhanced by artificial technology, allowing computers to carry out tasks that could be performed by man and even better than humans. In the artificial intelligence technique, computer systems are designed to learn and improve on their own based on available data.
Since the accurate diagnosis of malignant lesions is very crucial for better patient care, it would be quite important to explore the potential of artificial intelligence for the diagnosis of hematological malignant tumors.
Leukemias are quite common hematological malignancies which are characterized by the presence of neoplastic white blood cells in the bone marrow and sometimes in the peripheral blood. There are four main subtypes of leukemias which include acute lymphoblastic leukemia, acute myeloid leukemia, chronic lymphoid leukemia or chronic myeloid leukemia.^1-3
Acute leukemias are more frequently severe disorders in which the hemopoietic stem cell or early progenitors undergo malignant transformation. Acute leukemia is distinguished by having more than 20% blast cells in the blood or bone marrow at the time of clinical manifestation.^4-7 Less than 20% of blasts can be used to diagnose leukemia if specific cytogenetic or molecular genetic abnormalities are present.⁸
Acute lymphoblastic leukemia (ALL) is prevalent in children. Numerous nations across the world have observed an increase in the incidence of ALL.⁹ The accurate and early diagnosis of acute lymphoblastic leukemia is vitally important for their proper management. This category of cancer is diagnosed by identifying the blasts in the blood or bone marrow which could be done by the microscopic examination of the stained smears. The rising number of cases of acute lymphoblastic leukemia around the globe and the scarcity of highly qualified and trained health professionals in this field raise concerns regarding the accurate and prompt diagnosis of such cases for the proper treatment.
The advancement in digital technology has made it possible to develop machine learning systems with the help of neural networks for the evaluation of digitized microscopic images. The advancement in the development of high-resolution image data has made it quite feasible and achievable to make algorithms that use machine learning to extract the characteristic features of leukemia cells.
This study aims to investigate algorithms using artificial intelligence for the evaluation of digital image data to diagnose acute lymphoblastic leukemia.

Materials and Methods

The present study has been performed by acquiring one thousand one hundred fifteen anonymized digital photomicrographs from the Giemsa stained smears which included 515 digitized images of acute lymphoblastic leukemia and 600 images of normal peripheral blood cells. The digital images were labeled by four pathologists into two respective categories labeled as ALL and Normal. These anonymized digital images were subdivided into three sets. The first was a training set that included 360 digital images from the normal category and 309 photomicrographs from ALL category. The second was the validation set which included 120 images from the normal category and 103 images from ALL. The third was the test set. It contained 120 images from the normal category and 103 images from ALL.
To improve the accuracy of the models, fast ai and VGG16, ResNet18, and ResNet34 models were applied along with image augmentation techniques with the following details: Multiply=1, do flip=True, flip vert=True, max rotate=10, min zoom=1, max zoom=1.1, max lighting=0.2, max warp=0.2, p affine=0.75, p lighting=0.75, mode=“bilinear”, pad mode=“reflection”, align corners=True, min scale=1.Ethical ApprovalThis study was approved by the Ethics Committee of Northern Border University (Date: 26.04.2023, Decision No: 29/44/H) ( Local Committee of Bioethics Approval no. HAP-09-A-043).Statistical AnalysisThe obtained data from the application of algorithms were analyzed by calculating the parameters of specificity, sensitivity, positive predictive value, negative predictive value, and F1 scores.

Results

The ResNet18 and ResNet34 models obtained 100% accuracy for categories of acute lymphoblastic leukemia and normal blood cells on the test data. This model detected all 103 cases of acute lymphoblastic leukemia (ALL) without any mistake. Similarly, both models correctly detected all 120 cases of normal white blood cells. Both models had an F1-score of 1.0. The results are shown in Table 1.
The VGG16 model obtained an accuracy of 99% for ALL classification on the test data. With this model, 102 out of 103 digital images of acute lymphoblastic leukemia (ALL) were accurately diagnosed and 117 out of 120 images from normal blood cells were correctly identified. With the VGG16 model, the F1-score is 0.98.

Discussion

In the present study, the computer vision-based system revealed excellent results. It was created by using deep machine learning techniques to evaluate the digital images of normal white blood cells and neoplastic white blood cells. Three deep machine learning models were employed to diagnose acute lymphoblastic leukemia on the digitized images. These models include ResNet18, ResNet34, and VGG16. Based on the morphological features of neoplastic white blood cells, these deep learning algorithms can distinguish between images of acute lymphoblastic leukemia cells and normal white blood cells quite accurately. The higher F1 score of the ResNet18, ResNet34, and VGG16 algorithms indicated that pathologists could find the computer vision-based algorithms particularly useful in differentiating blood films. The findings of the present research are consistent with the findings published by Khandekar R et al.¹⁰
In another study, for the accurate differentiation between abnormal lymphoid cells and normal lymphocytes on the bone marrow specimen, the machine learning system revealed a positive predictive value of 99.04%.¹¹
In another published series, ResNet-50 and VGG-16 deep learning networks were applied to diagnose acute lymphoblastic leukemia and the validation accuracies of these were 81.6% and 84.6%, respectively, which are lower than the present study.¹²
Similarly, recent studies that employed artificial intelligence technology on whole slide images for the detection of morphological abnormalities related to malignancies revealed encouraging results.¹³
The application of deep machine learning technologies in this work yielded very encouraging results, reliably discriminating blood cells that are normal from acute lymphoblastic leukemia cells. The multiplicity of cells, their size, the amount of cytoplasm, and the size and shape of the nucleus in acute lymphoblastic leukemia cells may be the basis for their differentiation. This measurement and evaluation of these cell parameters necessitates specialized techniques and expertise in this area. There is a scarcity of hematopathology specialists. In this regard, the use of artificial technology as a tool could be quite beneficial.
Digital technology has accelerated task completion while also improving accuracy and cost-effectiveness. It has also decreased the possibility of mistakes that could have disastrous effects on human health. The process of automation and use of computer systems with the application of artificial intelligence could be an alternative technique for the microscopic evaluation of peripheral blood and bone marrow smears for the diagnosis of acute lymphoblastic leukemia.
The development of automated systems with intelligence-based algorithms may be an adjunct technique for the assistance of health professionals.^14-17 Artificial intelligence has been employed in speech recognition, radiological image analysis, and histopathological image evaluation.^18-21
Additional research using different algorithms will create opportunities for the use of AI technology in this field.

Limitations

Furthermore, the technique must be retested for differentiation of acute lymphoblastic leukemia subclasses L1, L2, and L3 to ensure its ability to differentiate the three subclasses.

Conclusion

The application of artificial intelligence for the diagnosis of acute lymphoblastic leukemia revealed encouraging results. In terms of accuracy & F1-score, the ResNet18 and ResNet34 models surpassed the VGG16 model. These models have the potential to be employed for better patient care in the near future.

Declarations

Abbreviations

References

1. Feng Q, de Smith AJ, Vergara-Lluri M, et al. Trends in acute lymphoblastic leukemia incidence in the United States by race/ethnicity from 2000 to 2016. Am J Epidemiol. 2021;190(4):519-527. doi:10.1093/aje/kwaa215
   Article · PubMed · Google Scholar
2. Kakaje A, Alhalabi MM, Ghareeb A, et al. Rates and trends of childhood acute lymphoblastic leukemia: an epidemiology study. Sci Rep. 2020;10(1):6756. doi:10.1038/s41598-020-63528-0
   Article · PubMed · Google Scholar
3. Guru Murthy GS, Pondaiah SK, Abedin S, Atallah E. Incidence and survival of T-cell acute lymphoblastic leukemia in the United States. Leuk Lymphoma. 2019;60(5):1171-1178. doi:10.1080/10428194.2018.1522442
   Article · PubMed · Google Scholar
4. Chiaretti S, Zini G, Bassan R. Diagnosis and subclassification of acute lymphoblastic leukemia. Mediterr J Hematol Infect Dis. 2014;6(1). doi:10.4084/mjhid.2014.073
   Article · PubMed · Google Scholar
5. Abbasi S, Maleha F, Shobaki M. Acute lymphoblastic leukemia experience: epidemiology and outcome of 2 different regimens. Mediterr J Hematol Infect Dis. 2013;5(1):e67-e492. doi:10.4084/mjhid.2013.024
   Article · PubMed · Google Scholar
6. Arber DA. The 2016 WHO classification of acute myeloid leukemia: what the practicing clinician needs to know. Semin Hematol. 2019;56(2):90-95. doi:10.1053/j.seminhematol.2018.08.002
   Article · PubMed · Google Scholar
7. Dong Y, Shi O, Zeng Q, Lu X, Wang W, Li Y, Wang Q. Leukemia incidence trends at the global, regional, and national level between 1990 and 2017. Exp Hematol Oncol. 2020;9:14. doi:10.1186/s40164-020-00170-6
   Article · PubMed · Google Scholar
8. El Achi H, Khoury JD. Artificial intelligence and digital microscopy applications in diagnostic hematopathology. Cancers (Basel). 2020;12(4):797. doi:10.3390/cancers12040797
   Article · PubMed · Google Scholar
9. Yi M, Zhou L, Li A, Luo S, Wu K. Global burden and trend of acute lymphoblastic leukemia from 1990 to 2017. Aging (Albany NY). 2020;12(22):22869-22891. doi:10.18632/aging.103982
   Article · PubMed · Google Scholar
10. Khandekar R, Shastry P, Jaishankar S, Faust O, Sampathila N. Automated blast cell detection for acute lymphoblastic leukemia diagnosis. Biomed Signal Process Control. 2021;68:102690. doi:10.1016/j.bspc.2021.102690
    Article · PubMed · Google Scholar
11. Tang G, Fu X, Wang Z, Chen M. A machine learning tool using digital microscopy for the identification of abnormal lymphocytes in the bone marrow. Acta Cytol. 2021;65(4):354-357. doi:10.1159/000518382
    Article · PubMed · Google Scholar
12. Rezayi S, Mohammadzadeh N, Bouraghi H, Saeedi S, Mohammadpour A. Timely diagnosis of acute lymphoblastic leukemia using artificial intelligence-oriented deep learning methods. Comput Intell Neurosci. 2021;2021:5478157. doi:10.1155/2021/5478157
    Article · PubMed · Google Scholar
13. El Achi H, Khoury JD. Artificial intelligence and digital microscopy applications in diagnostic hematopathology. Cancers (Basel). 2020;12(4):797.
    PubMed · Google Scholar
14. Eckardt JN, Bornhäuser M, Wendt K, Middeke JM. Application of machine learning in the management of acute myeloid leukemia: current practice and future prospects. Blood Adv. 2020;4(23):6077-6085. doi:10.1182/bloodadvances.2020002997
    Article · PubMed · Google Scholar
15. Bukhari SUK, Syed A, Bokhari SKA, Shah SSH. The efficiency of deep learning for the diagnosis of psammomatous meningioma. Ann Clin Anal Med. 2021;12(2):153-156. doi:10.4328/acam.20226
    Article · PubMed · Google Scholar
16. Rehman A, Abbas N, Saba T, Rahman SIU, Mehmood Z, Kolivand H. Classification of acute lymphoblastic leukemia using deep learning. Microsc Res Tech. 2018;81(11):1310-1317. doi:10.1002/jemt.23139
    Article · PubMed · Google Scholar
17. Nikitaev VG, Tupitsyn NN, Pronichev AN, et al. Artificial intelligence technologies in the diagnosis of acute lymphoblastic leukemia and minimal residual disease. Biomed Eng. 2021;54:354-356. doi:10.1007/s10527-021-10038-6
    Article · PubMed · Google Scholar
18. Lee JG, Jun S, Cho YW, et al. Deep learning in medical imaging: general overview. Korean J Radiol. 2017;18(4):570-584. doi:10.3348/kjr.2017.18.4.570
    Article · PubMed · Google Scholar
19. Pałczyński K, Śmigiel S, Gackowska M, et al. IoT application of transfer learning in hybrid artificial intelligence systems for acute lymphoblastic leukemia classification. Sensors (Basel). 2021;21(23):8025. doi:10.3390/s21238025
    Article · PubMed · Google Scholar
20. El Hussein S, Chen P, Medeiros LJ, et al. Artificial intelligence strategy integrating morphologic and architectural biomarkers provides robust diagnostic accuracy for disease progression in chronic lymphocytic leukemia. J Pathol. 2022;256(1):4-14. doi:10.1002/path.5795
    Article · PubMed · Google Scholar
21. El Alaoui Y, Elomri A, Qaraqe M, et al. A review of artificial intelligence applications in hematology management: current practices and future prospects. J Med Internet Res. 2022;24(7):520-531. doi:10.2196/36490
    Article · PubMed · Google Scholar

Tables

Additional Information

Publisher’s Note
Bayrakol MP remains neutral with regard to jurisdictional and institutional claims.

Rights and Permissions

Creative Commons License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). To view a copy of the license, visit https://creativecommons.org/licenses/by-nc/4.0/

View full license details →