AI to minimise human errors in the detection of hematological diseases ⋆ BanaFridath BioNigan Ecole Supérieure Multinationale des Télécommunications Laboratory E-Inov
Dakar SENEGAL
AlbanGildas Zohoun Faculté des Sciences de Santé Laboratoire d'hématologie CNHU-HKMBENIN
Cotonou
AhmedDooguy Ecole Supérieure Multinationale des Télécommunications Laboratory E-Inov
Dakar SENEGAL
AI to minimise human errors in the detection of hematological diseases ⋆ 1613-0073 4832638785AC6C0B9E62F56C71698515 GROBID - A machine learning software for extracting information from scholarly documents AI errors hematology minimise

Hematology is a branch of medicine that relies on accurate diagnosis and appropriate treatment of blood-related diseases. However, human errors, whether due to technician fatigue, inattention or technical limitations, can have serious consequences for patients. Artificial intelligence offers a solution to these problems. By integrating advanced machine learning and deep learning algorithms, AI offers innovative solutions for reducing diagnostic and treatment errors. With its ability to analyse big data with high accuracy, AI promises to transform hematology practice, ensuring safer and more effective care for patients. This article reviews the different AI techniques used in the recognition of blood cells and the detection of related diseases, while highlighting its benefits in minimising human errors in diagnosis.
Introduction

Artificial intelligence (AI) is defined as a process of imitating human intelligence, based on the creation and application of algorithms executed in a dynamic computer environment [1]. It is therefore the ability of a machine to imitate human behavior (analysis, interpretation, decision-making based on an image) via algorithms and to make predictions based on data already acquired [2]. AI is faster and makes fewer errors than humans when it comes to achieving results. Its importance is growing by the day. It has developed and is now present in almost every sector of human activity: transport, agriculture, commerce, medicine, .... The use of AI in medicine began in the 20th century in developed countries for the rapid management of patients and the accurate diagnosis of certain diseases [3]. Today, AI is commonly used in the detection of rare genetic diseases (Cornelia de Lange syndrome, Angelman syndrome, etc.) [4], heart disease and cancer [5], blood diseases, …. It is transforming many areas of medicine, including hematology. Hematology is a medical speciality that studies blood, the hematopoietic organs (bone marrow, lymph nodes and spleen being the main ones) and their diseases [6]. Blood includes blood cells in plasma which are made in the red bone marrow from a stem cell. By dividing and differentiating, this cell gives rise to one of three categories of blood cells [7]:

• Red blood cells (RBC), also known as erythrocytes, which are anucleate cells and are the most numerous, around 5 million/mm3. • Platelets (PLT), which are anucleate fragments and occur at a rate of 150,000 to 450,000/mm3.

Each type of cell has its own distinguishing features, whether in terms of shape, colour or even size.

In most of our hematology laboratories in Africa and in Benin in particular (CNHU-HKM), our technicians carry out this recognition work manually using a microscope. In addition to the long wait by patients before obtaining results, these results are sometimes exposed to human errors of inattention. This is justified by the large number of blood slides to be analysed by these technicians. By using advanced algorithms and massive data processing capabilities, AI offers unprecedented opportunities to improve the diagnosis, treatment and management of hematological diseases and sometimes reduce the cost of analysis. This is the case of an AI model used by some authors to help practitioners identifying different hematological diseases with inexpensive hemogram tests. This binary and multiclass classification model achieved up to 96% accuracy [8]. This article explores the different applications of AI in hematology, highlighting the potential benefits of this revolutionary technology. We first present the litterature review on recognition and classification of blood cells and automatic detection of blood diseases. Then, we present the CNN model designed for the CNHU-HKM hematology laboratory and performances obtained. Finally, we discuss the benefits of AI in minimising human error in hematology diagnosis.

Litterature Review
Automatic detection of blood cells diseases 2.2.1. White blood cells

The excessive presence of certain immature cells in the peripheral blood reveals that the patient has a disease, the case of blasts for leukemia [27].

Red blood cells

The literature is full of studies on the detection of diseases caused by RBCs.

Performances of the proposed model for the CNHU-HKM hematology laboratory

The model designed for CNHU-HKM hematology laboratory is a CNN architecture of 12-layers. It allows to recognize blood cells (WBC, RBC) and detect blood cells diseases like sickle cell anemia in CNHU-HKM hematology laboratory [44]. These layers are optimized for the maximum positive prediction rate. The loss curve evaluates how well our algorithm models the dataset. The lower the loss, the better is. Figure3 and 4 show performance obtained in the automatic classification of blood cells for 32*32 images and 16*16 images respectively. We notice that while the training curve tends towards zero (0), the validation curve is a bit high [45]. This can help diagnose hematological diseases more quickly and accurately. • Clinical decision support: AI systems can provide recommendations based on clinical data and best practice, helping doctors to make more informed decisions. For example, in hematology, AI can suggest optimal treatments for diseases such as leukemia or anemia. • corresponding author mark :

\cormark[<num>]Automation of repetitive tasks: AI can automate administrative and clinical tasks, reducing the risk of human error. For example, the transcription of test results or the management of medical records can be carried out by AI systems, which relieves healthcare professionals.

• Continuous monitoring: AI systems can continuously monitor patients and alert doctors to significant changes in health parameters. This is particularly useful for patients with chronic diseases or who require constant monitoring, such as certain hematological diseases.

Conclusion

In conclusion, AI integration in hematology represents a major advance that promises to transform clinical and research practices. Although challenges remain, particularly in terms of regulation, ethics and acceptance by healthcare professionals, the potential benefits of AI are immense. By improving diagnostic accuracy, optimizing treatments, and facilitating data management, AI has the potential to reduce human error and improve patient outcomes. The future of hematology, enriched by AI, looks promising and will bring significant innovations.

Normal and abnormal cells were classified into four classes: sickle cells, dacrocytes, ovalocytes and erythrocytes by Sharma V (2016) using the KNN classifier and Watershed segmentation technique with an accuracy of 80.6%[36].Xu M et al. (2017) focused on the RBC shape detection using different techniques. A Deep CNN was used to find their region of interest (ROI) using an automatic seed generation technique and a mask based on patch normalization to obtain images of uniform size. This method is not widely used because it requires a very large database[37]. Sobel's edge detection algorithm is used by Mohamad A et al. (2017) for detecting RBC shape with blob measurement. This inexpensive technique is beneficial for people living in remote areas and achieveded 95% accuracy but only for 2D images[38]. Zhang M et al. (2020) adopted a semantic segmentation framework based on deep learning to solve the GR classification task. The performance obtained was 97% for the dU-Net model and 94.7% for the classical U-Net model[39]. A transfer learning technique that automatically extracts features and is specific to small databases has also been proposed byAlzubaidi L et al. (2020). Thanks to data augmentation, it achieved 99.98%[40]. Chy T et al. (2019) used different techniques (fuzzy C mean clustering algorithm, KNN, SVM and ELM) to automatically detect sickle cell disease. ELM classifier performed better, with an accuracy of 95.45%[41]. AlexNet was also used byAliyu H. et al. (2020) to classify red blood cells in sickle cell anemia. The accuracy obtained was 95.92%[42]. Another technique using a smartphone microscope has also been used on blood smears for the same purpose by De Haan K. et al. (2020). It comprises two distinct and complementary deep neural networks and achieved an accuracy of around 98%[43].
Figure 1 :Figure 1: Accuracy curve 32*32
Figure 2 :Figure 2: Accuracy curve 16*16
Figure 3 :Figure 3: Loss curve 32*32
Figure 4 :Figure 4: Loss curve 16*16
Figure 5 :Figure 5: Accuracy curve 32*32
Figure 6 :Figure 6: Accuracy curve 16*16
2.1. Recognition and Classification of Blood Cells 2.1.1. White blood cells (WBC)RB Hegde et al. (2019) also compared traditional and DL methods and arrived at the same results with 99% of accuracy for WBCs classification[10]. ASAshour et al. (2021) used a neural network associated with the SVM algorithm on a database combination and achieved a segmentation performance of 94.9counting accuracy for both cells is 97.4Ruberto et al. (2015) created a multi-classifier system for WBC segmentation using ANN + Nearest Neighbor and SVM algorithms and obtained a segmentation accuracy of 99%[12]. S Manik et al. (2016) used an ANN-based system with MAT-LAB capabilities to detect and classify WBC. The accuracy of the entire system is 98.9% with 100for eosinophils and neutrophils, 96.7% for lymphocytes[13]. Using the Faster R-CNN (Fast R-CNN + Region Proposal Network (RPN)) method on the BCCD dataset to recognize and classify different blood cells, S Raina et al. (2020) obtained the following results: RBC 55.83%, PLT 68.36% and WBC 92.10% [14]. J Basnet et al. (2020), using a Deep CNN-based method, improved classification accuracy from 96.1% to 98.92% andFor cell recognition, authors used different ML/DL methodsfor cell segmentation, classification, and counting.
S Khan et al. (2021) used both traditional learning methods (manual extraction of features + classification cells by ANN) and DL-based methods (characteristic extraction + classification cells by CNN) to classify WBCs. This study reveals that they achieved the same performance in all 02cases[9]. reduced processing time from 0.354 to 0.216 s[15]. MJ Macawile et al. (2018) created a CNN-based system to classify and count WBCs using the HSV (Hue Saturation Value) saturation component on the ALL-IDB database. They compared several models (Alexnet, GoogleNet, and ResNet-101). AlexNet appears to be the best, with a sensitivity of 89.18%, a specificity of 97.85% and an accuracy of 96.63%[16]. With a model based on the Deep RN and using the characteristics of the convolutional layers of the AlexNet architecture, AKhan et al. (2021) were able to identify the different types of WBCs with a training accuracy of 99.99% and the test of 99.12%. This model is named MLANet-FS-ELM[17]. A Malkawi et al. (2020) set up a CNN-based hybrid system to extract WBCs characteristics and classify them. They evaluated the performance of 3 classifiers (SVM, k-NN, RF) on the LISC WBC and the RF performed better with a test accuracy of 98.7%[18]. A. Şengür et al. (2019) used a system based on image processing and ML in particular Deep CNN to classify WBCs according to their shape and other deep characteristics. This system achieved an accuracy of 80% in relation to the shape and 82.9% in relation to the deep characteristics; combining these 02 parameters, the overall accuracy is 85.7 2.
1.2. Red blood cells, echinocytes, lacrimal cells and macrocytes) with an accuracy of 91.667% over a period of 0.81432 seconds for different blood samples[25]. Namata et al (2021) proposed an image processing method using a convolution neural network for classification of RBCs. The algorithm used extracts features from segmented images and classifies in nine categories with an overall accuracy of 98.5%[26].Differents ML/DL techniques have been used for automaticcell recognition [20].Maity et al. (2012) employed an efficient supervised-decision-tree C4.5 to classify RBCs into six sub-classes in-cluding sickle-cells with 98.2% precision and 99.6% speci-ficity [21]. They also proposed another method which em-phasizes the extraction of crucial shape-based features forRBC classification into nine classes including healthy cellsin 2017. This method achieved 99.71% specificity and 97.81%accuracy [22].Acharya and Kumar (2017) employed a technique capable toclassify RBCs into 11 sub-classes including sickle-cells with98% precision [23]. Using the Hough Circular Transform
(HCT) method,Mazalan SM et al. (2013) were able to count the total number of RBCs in a peripheral blood smear image. Results showed that from ten sample peripheral blood smear images, accuracy was 91.87%[24]. Using the same method, Chadha GK et al. (2020) were able to count and classify RBCs according to four types of abnormality (ellip-tocytes
Figure1 and 2 show performance obtained in automatic classification of blood cells for 32*32 images and 16*16 images respectively with accuracy of 98.78% for training and 88.11% for test and 86.59% for cross validation for 32*32 images, and 90.11% for test and 88.53% for cross validation with 16*16 images[44].
Acknowledgments

We thank Professor Issiako Bio Nigan for his recommendations. We also thank the CNHU-HKM hematology laboratory of Benin for giving us access to data.

Intelligence artificielle: définition et utilisations 2023 Parlement européen Unesco Qu'estce-que l'intelligence artificielle 2023 Site officiel de la Chaire UNESCO De l'intelligence artificielle à son application en médecine PCallier OSandel 2021 Actualités pharmaceutiques Une intelligence artificielle qui détecte les maladies génétiques rares PLabbe Sama game -l'arène des nouvelles sur les jeux Français PHughes 2023 MarionSpée Hématologie 2016 SWeb circule dans les vaisseaux sanguins Le sang (hématies, plaquettes SergioDiaz-Del-Pino RobertoTrelles-Martinez FAGonzález-Fernández NicolasGuil 10.1016/j.he-liyon.2023.e15940 Artificial intelligence to assist specialists in the detection of haematological diseases Elsevier Ltd 2023 A Review on Traditional Machine Learning and Deep Learning Models for WBCs Classification in Blood Smear Images SKhan MSajjad THussain AUllah ASImran 10.1109/AC-CESS.2020.3048172 IEEE Access 9 2021 Comparison of traditional image processing and deep learning approaches for classification of white blood cells in peripheral blood smear images RBHegde KPrasad HHebbar BM KSingh 10.1016/j.bbe.2019.01.005 Biocybernetics and Biomedical Engineering 39 2 2019 A Cascaded Classification-Segmentation Reversible System for Computer-Aided Detection and Cells Counting in Microscopic Peripheral Blood Smear Basophils and Eosinophils Images ASAshour MAWahba RGhannam 10.1109/AC-CESS.2021.3083703 IEEE Access 9 2021 A Multiple Classifier Learning by Sampling System for White Blood Cells Segmentation CDi Ruberto ALoddo LPutzu 10.1007/978-3-319-23117-4-36 Computer Analysis of Images and Patterns Lecture Notes in Computer Science GAzzopardi NPetkov
Cham
 Springer International Publishing 2015 9257 Counting and classification of white blood cell using Artificial Neural Network (ANN) SManik LMSaini NVadera 10.1109/ICPEICES.2016.785364 1st International Conference on Power Electronics, Intelligent Control and Energy Systems (ICPEICES)
Delhi, India
 IEEE 2016 Blood Cells Detection Using Faster-RCNN SRaina AKhandelwal SGupta ALeekha 10.1109/GU-CON48875.2020.9231134 International Conference on Computing, Power and Communication Technologies (GUCON)
Greater Noida, India
 IEEE 2020 A Novel Solution of Using Deep Learning for White Blood Cells Classification: Enhanced Loss Function with Regularization and Weighted Loss (ELFRWL) JBasnet AAlsadoon PW CPrasad SAAloussi OHAlsadoon 10.1007/s11063-020-10321-9 Neural Process Lett 52 2 2020 White blood cell classification and counting using convolutional neural network MJMacawile VVQuinones ABallado JDCruz MVCaya 10.1109/ICCRE.2018.8376476 3rd International Conference on Control and Robotics Engineering (ICCRE)
Nagoya
 IEEE 2018 White blood cell type identification using multilayer convolutional features with an extremelearning machine AKhan AEker AChefranov HDemirel 10.1016/j.bspc.2021.102932 Biomedical Signal Processing and Control 69 102932 2021 White Blood Cells Classification Using Convolutional Neural Network Hybrid System AMalkawi RAl-Assi TSalameh BSheyab HAlquran AMAlqudah 10.1109/MECBME47393.2020.9265154 5th Middle East and Africa Conference on Biomedical Engineering (MECBME)
Amman, Jordan
 IEEE 2020 White Blood Cell Classification Based on Shape and Deep Features ASengur YAkbulut UBudak ZComert 10.1109/IDAP.2019.8875945 International Artificial Intelligence and Data Processing Symposium (IDAP)
Malatya, Turkey
 IEEE 2019 Analyses de laboratoire pour les maladies du sang MsdManuels Computerassisted approach to anemic erythrocyte classification using blood pathological information MMaity PSarkar CChakraborty Proc. 3rd Int. Conf. Emerg. Appl. Inform. Technol 3rd Int. Conf. Emerg. Appl. Inform. Technol 2012 An ensemble rule learning approach for automated morphological classification of erythrocytes MMaity TMungle DDhane AKMaiti CChakraborty J. Med. Syst 41 2017 Identification and red blood cell classification using computer aided system to diagnose blood disorders VAcharya PKumar Proc. Int. Conf. adv. Comput., Inform Int. Conf. adv. Comput., Inform 2017 Automated red blood cells counting in peripheral blood smear image using circular hough transform SMMazalan NHMahmood MaaRazak Dir 1st International Conference on Artificial Intelligence, Modelling and Simulation <author> <persName><forename type="first">Kota</forename><surname>Kinabalu</surname></persName> </author> <author> <persName><forename type="first">Malaisie</forename></persName> </author> <idno type="DOI">10.1109/AIMS.2013.59</idno> </analytic> <monogr> <title level="j">IEEE 2013 An automated method for counting red blood cells using image processing GKChadha Srivastavaa ASingh RGupta DSingla 10.1016/j.procs.2020.03.408 Procedia Computer Science 167 2020 Red Blood Cell Classifcation Using Image Processing and CNN APMamata DMNinad 10.1007/s42979-021-00458-2 SN Computer Science 2 70 2021 SantéSm Magazine Avail- able Leucémie : symptômes, diagnostic, traitements 2021 White Blood Cells Classification with Deep Convolutional Neural Networks MJiang LCheng FQin LDu MZhang 10.1142/S0218001418570069 Int. J. Patt. Recogn. Artif. Intell 32 09 1857006 2018 Advanced machine learning for leukaemia detection based on white blood cell segmentation IMSheikh Chachoo Ma 10.1007/978-981-15-9671-1-17 Information, Communication and Computing Technology CBadica PLiatsis LKharb DChahal
Singapore; Singapore
 Springer 2020 A convolutional neural network-based learning approach to acute lymphoblastic leukaemia detection with automated feature extraction SAnwar AAlam 10.1007/s11517-020-02282-x Med Biol Eng Comput 58 12 2020 A deep learning model (ALNet) for the diagnosis of acute leukaemia lineage using peripheral blood cell images LBoldú AMerino AAcevedo AMolina JRodellar 10.1016/j.cmpb.2021.105999 Computer Methods Programs Biomed 202 105999 2021 Automated detection of white blood cells cancer diseases HMohamed ROma NSaeed AEssam NAyman AMoyiy First International Workshop on Deep and Representation Learning IWDRL <idno type="DOI">10.1109/IW-DRL.2018.8358214</idno> <imprint> <date type="published" when="2018">2018</date> <publisher>IEEE</publisher> <pubPlace>Cairo; Cairo</pubPlace> </imprint> </monogr> </biblStruct> <biblStruct xml:id="b34"> <analytic> <title level="a" type="main">A blood cell dataset for lymphoma classification using faster R-CNN BSheng MZhou MHu QLi LSun YWen 10.1080/13102818.2020.1765871 Biotechnol Biotechnol Equipment 34 1 2020 Detection of white blood cell cancer using image processing RAgrawal SSatapathy GBagla KRajakumar Dir 10.1109/ViTE-CoN.2019.8899602 International Conference on Vision Towards Emerging Trends in Communication and Networking (ViTECoN)
Vellore, India; Vellore
 IEEE mars 2019. 2019 A decision support system for Acute Leukaemia classification based on digital microscopic images ASNegm OAHassan AHKandil 10.1016/j.aej.2017.08.025 Alexandria Engineering J 57 4 2018 Detection of sickle cell anemia and thalassemia causing abnormalities in thin smear of human blood sample using image processing VSharma ARathore GVyas International Conference on Invention Computation Technologies (ICICT) 26 7 <author> <persName><forename type="first">India</forename><surname>Coimbatoire</surname></persName> </author> <idno type="DOI">10.1109/INVENTIVE.2016.7830136</idno> </analytic> <monogr> <title level="j">Coimbatoire: Med 2016 A deep convolutional neural network for classification of red blood cells in sickle cell anemia MXu DPPapageorgiou SZAbidi MDao HZhao GEKarniadakis 10.1371/journal.pcbi.1005746 PLoS Comput Biol 13 10 e1005746 2017 Sickle cell disease verification via sobel edge algorithms for image processing AMohamad RHamzah AMokhtar JSathar Dir International Conference on Engineering Engineering Technology and Technopreneurship <author> <persName><surname>Malaisie</surname></persName> </author> <idno type="DOI">10.1109/ICE2T.2017.8215994</idno> </analytic> <monogr> <title level="j">Research Gate 2017 Automated semantic segmentation of red blood cells for sickle cell disease MZhang XLi MXu QLi 10.1109/JBHI.2020.3000484 IEEE J Biomed. Health Inform 24 11 2020 Deep learning models for classification of red blood cells in microscopy images to aid in sickle cell anemia diagnosis LAlzubaidi MAFadhel OAl-Shamma JZhang YDuan 10.3390/electronics9030427 Electronics 9 3 427 2020 A comparative analysis by KNN, SVM, ELM classification to detect sickle cell anemia TChy MRahaman International Conference on Robotics, Electrical and Signal Processing Techniques ICREST <idno type="DOI">10.1109/ICREST.2019.8644410</idno> <imprint> <date type="published" when="2019">2019</date> <publisher>IEEE</publisher> <pubPlace>Dhaka, Bangladesh; Dhaka</pubPlace> </imprint> </monogr> </biblStruct> <biblStruct xml:id="b46"> <analytic> <title level="a" type="main">A deep learning AlexNet model for classification of red blood cells in sickle cell anemia HAliyu MRazak RSudirman NRamli 10.11591/ijai.v9.i2.pp221-228 IAES IJ-AI 9 2 2020 Automated screening of sickle cells using a smartphone-based microscope and deep learning KDe Haan CeylanKoydemir HRivenson YTseng DVan Dyne EBakic L 10.1038/s41746-020-0282-y Digit Med 3 1 76 2020 White Blood Cells Recognition and Classification using Convolutional Neural Network BFBio Nigan AGZohoun ADKora 10.1109/ICAAIC56838.2023.10140293 2nd International Conference on Applied Artificial Intelligence and Computing (ICAAIC) 2023 Automatic Detection of Sickle Cell Disease, Elliptocytosis and Schizocytosis BFBio Nigan AGZohoun ADKora 10.1109/ICASI60819.2024.10547902 10th International Conference on Applied System Innovation (ICASI) 2024