Introduction
Hematological diseases, also known as blood system diseases, refer to diseases that originate in or mainly involve the blood and hematopoietic organs and can be found in people of different ages [ 1 ]. The common diseases include anemia, bleeding, and fever [ 2 ], and their causes include genetic defects and drug damage. The main types are iron-deficiency anemia, aplastic anemia, leukemia, and immune thrombocytopenic purpura [ 3 , 4 ]. Early diagnosis and treatment of this disease are crucial, and therefore improving awareness of hematological diseases is crucial [ 5 – 7 ].
Pulmonary fungal infections (PFI) involve fungal inflammation or related lesions in the lungs and bronchi. Symptoms of PFI can be mild or severe, depending on the pathogen and host status [ 8 , 9 , 10 ]. Blood diseases can lead to weakened immunity and abnormal blood count in patients, thereby increasing the risk of PFI, and causes patients to experience symptoms such as fever, cough, drowsiness, limb weakness, and loss of appetite [ 11 ]. Clinical impact assessment of PFI commonly involves chest X-ray examination [ 12 ]. Computed tomography (CT) is the most common X-ray imaging technology. With shadows and halo signs visible in the early stages of infection, 10–15 days after the onset, liquefaction, and necrosis of the pulmonary consolidation region are shown in CT, which has a good diagnostic performance [ 13 ].
Currently, radiomics can transform images into digital features through machine fitting. In recent years, artificial intelligence algorithms have made significant progress in the field of medical image diagnostics and have been widely applied to the diagnosis and assessment of pulmonary diseases [ 14 – 16 ]. Secondary PFIs in hematological malignancies (HM) can significantly impact patients’ quality of life and prognosis. However, traditional radiological methodologies have certain limitations in accurately diagnosing fungal infections in the lungs. Definitive determination often necessitates further sampling and examination by medical professionals. In recent years, the application of machine algorithms for assisting medical diagnosis has gained widespread acceptance, and the utilization of artificial intelligence-based CT imaging techniques has shown tremendous potential. This research primarily investigated the utility of AI-based CT imaging in the diagnosis of secondary PFIs in patients with HM. The support vector machine (SVM) algorithm has been widely applied in medical imaging, as indicated by previous studies [ 17 , 18 ]. In the context of diagnosing secondary PFIs in HM, SVM demonstrates the capability for precise localization of lesions. Through large-scale sample training, SVM algorithm models achieve higher diagnostic accuracy, effectively distinguishing trained disease images from those of other conditions. This aids in facilitating precise treatment decisions [ 19 ]. Furthermore, CT imaging based on the SVM algorithm can expedite the diagnostic process within a relatively short time frame. Secondary PFIs in the context of HM represent a potentially dangerous and severe complication. Leveraging CT imaging technology based on the SVM algorithm offers distinct advantages, particularly in the early detection and treatment of PFIs. This not only enhances the efficiency of healthcare professionals but also ensures patients receive more timely treatment, thus providing robust support for clinical decision-making.
In this study, CT examinations were conducted on a cohort of 100 patients suffering from HM complicated by secondary PFI. Utilizing CT images, an analysis of the pathogens and radiological features observed in different PFI patients was conducted. Subsequently, features from the CT images were extracted using the SVM algorithm. This algorithm was then trained on data from the training set of patients to enhance its diagnostic accuracy. The ultimate goal of this study was to provide valuable insights for the clinical diagnosis and treatment of secondary PFIs in patients with HM. Therefore, this research holds significant application value and broader implications in the field of biology.
Literature review
Godoy et al. [ 20 ] highlighted the susceptibility of patients with HM and hematopoietic stem cell transplant (HSCT) recipients to fungal pulmonary infections. The most common pathogens associated with these infections include Aspergillus , Mucor , Rhizopus , and Candida species. Due to the immunocompromised state or immunodeficiency in these patients, fungal pneumonia exhibits a significantly elevated incidence rate. In the research by Lewis et al. [ 21 ] focusing on the diagnosis of fungal pulmonary infections in patients with HM, it was emphasized that high-resolution chest CT remains the preferred diagnostic imaging modality for invasive fungal diseases. CT imaging can reveal features such as nodules with or without ground-glass opacities, masses, consolidations, wedge-shaped infarctions, and pleural effusions. Furthermore, with the development of fungal-specific antibody imaging tracers, CT has become an indispensable tool in the diagnosis of this condition. Kunihiro et al. [ 22 ] conducted CT examinations on 128 patients with pulmonary infections associated with HM. The infection types among the patients included bacterial pneumonia (37 cases in non-HSCT cases and 14 cases in HSCT cases), Pneumocystis pneumonia (PCP) (29 cases in non-HSCT cases and 11 cases in HSCT cases), and fungal infections other than PCP (20 cases in non-HSCT cases and 17 cases in HSCT cases). After evaluating the CT scans of the patients, the doctors conducted comparisons using chi-square tests and multivariate logistic regression analysis. They observed that CT imaging could distinctly depict features such as centrilobular nodules, nodules with a halo sign, and cystic lesions with air crescent signs in the patients. Furthermore, based on pathogen testing, it was found that patients with Aspergillus infections exhibited early CT findings of nodules with a halo sign, while in later stages, they displayed cystic lesions with air crescent signs. Patients with Cryptococcus infections exhibited CT imaging features characterized by nodules and cavities, while those with Candida infections presented with centrilobular nodules or randomly distributed small nodules. Hałaburda-Rola et al. [ 23 ] conducted CT examinations on 35 neutropenic patients with invasive pulmonary aspergillosis who had symptoms such as fever, dyspnea, or nonproductive cough occurring within 72 h. They applied the European Organization for Research and Treatment of Cancer/Mycoses Study Group (EORTC/MSG) criteria and found that CT’s diagnostic value increased by 11.4%. Martinez et al. [ 24 ] trained and tested various machine learning systems on 264 cases of invasive fungal diseases and 289 control patients. The results indicated that the SVM algorithm model exhibited a high recall rate at the patient level, reaching 95 and 100%, respectively. Qian et al. [ 25 ] included 40 patients with pulmonary aspergillosis, divided into a training group ( n = 28) and a testing group ( n = 12). They performed CT imaging, extracted radiomic features, and established a diagnostic model based on the SVM algorithm. The area under the curve (AUC) values for the training and testing sets were 0.896 and 0.886, respectively. These findings highlight the SVM algorithm’s remarkably high accuracy in diagnosing pulmonary aspergillosis. In summary, SVM algorithm models exhibit the highest sensitivity in analyzing CT images and can diagnose specific fungal infections based on radiomic features, thereby enhancing diagnostic efficiency.
Materials and methods
Research objects
One hundred patients with HM complicated by secondary PFI in the hospital from March 10, 2020, to February 10, 2022, were selected as the research objects, serving as training sets. Sixty-eight males and 32 females were included, with an age range of 10–65 years old. The patients in the test set were 80 patients with HM complicated with secondary PFI admitted to our hospital during the same period. Among them, there were 51 males and 29 females, with an age range of 12–70 years. This research had been approved by the ethics committee of the hospital, and the family members of the patients knew about the research and signed the informed consent.
CT imaging examination
The patients were examined by dual-source CT (Manufacturer: Siemens, Germany, Model: SOMATOM Drive) scanner. The patient was in a supine position with the arms raised above the head, and the head went ahead through the instrument. Scanning was from the superior aperture of the thorax to the mid-abdomen. Scanning parameters included a tube voltage of 120 kV, a tube current of 150 mA, a slice spacing of 5.5 mm, and a slice thickness of 5.5 mm. The obtained CT images were sent to the workstation for processing, and the images were read and diagnosed by two senior physicians together. The features of CT images were analyzed.
Image segmentation
At the critical level of image lesion characterization, the image window width and window level were set to 400 Hounsfield Units (Hu) and 50 Hu, respectively. Subsequently, employing ITK-SNAP 3.8 software, the diagnostic region of interest (ROI) was delineated along the contours of the lesion. The outlined ROI ( Figure 1 ) was then blindly reviewed and outlined by a radiologist from the hospital’s radiology department. The maximum slice image of the focused-enhanced CT along the transverse axis was imported, and corresponding ROI images were drawn into the A.K. software for feature extraction. A total of 24 features were extracted from the axial images of the maximum slice of each lesion, encompassing morphological features, first-order histogram features, second-order histogram features, and higher-order features.
Feature extraction
The research cohort comprised 100 patients with malignant hematological disorders complicated by secondary PFI (research group) and 80 patients with malignant hematological disorders as the control group, aimed at mitigating the influence of dimensionality and value range disparities. Initial feature selection in the training set was conducted using the minimum redundancy maximum relevance algorithm, which eliminated redundant and irrelevant features while retaining those with the highest predictive efficiency: mRMR = argmax f i ∈ X I ( c ; f i ) − 1 ∣ S ∣ ∑ f j ∈ S I ( f i ; f j ) . \documentclass[10pt]{article}\usepackage{wasysym}\usepackage[substack]{amsmath}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage[mathscr]{eucal}\usepackage{mathrsfs}\usepackage{pmc}\usepackage[Euler]{upgreek}\pagestyle{empty}\oddsidemargin -1.0in\begin{document}\[\text{mRMR}=\text{argmax}{f}_{i}\in X\left[I(c;{f}_{i})-1| S| \sum {f}_{j}\in SI({f}_{i};{f}_{j})\right].]\]\end{document}
The original data were transformed into linear data and mapped between 0 and 1, inclusive: NMIFS = argmax f i ∈ X I ( f i ; c ) − 1 ∣ S ∣ ∑ f j ∈ S I ( f i ; f j ) min ( H ( f i ) , H ( f j ) ) . \documentclass[10pt]{article}\usepackage{wasysym}\usepackage[substack]{amsmath}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage[mathscr]{eucal}\usepackage{mathrsfs}\usepackage{pmc}\usepackage[Euler]{upgreek}\pagestyle{empty}\oddsidemargin -1.0in\begin{document}\[\text{NMIFS}=\text{argmax}\hspace{.25em}{f}_{i}\in X\left[I({f}_{i}\text{;}c)-1| S| \sum {f}_{j}\in SI({f}_{i}\text{;}{f}_{j})\text{min}(H({f}_{i}),H({f}_{j}))\right].]\]\end{document}
Subsequently, the least absolute shrinkage and selection operator algorithm was employed for further feature selection, and the objective function of the prediction model was presented as follows: ∑ n i = 1 y i − ∑ p j = 1 x i j β j 2 + λ ∑ p j = 1 ∣ β j ∣ = RSS + λ ∑ p j = 1 ∣ β j ∣ . \documentclass[10pt]{article}\usepackage{wasysym}\usepackage[substack]{amsmath}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage[mathscr]{eucal}\usepackage{mathrsfs}\usepackage{pmc}\usepackage[Euler]{upgreek}\pagestyle{empty}\oddsidemargin -1.0in\begin{document}\[\sum {n}_{i}=1\left(\phantom{\rule[-0.75em]{}{0ex}}{y}_{i}-\sum {p}_{j}=1{x}_{ij}{\beta }_{\text{j}}\right)2+\lambda\sum {p}_{j}=1| {\beta }_{j}| =\text{RSS}+\lambda \sum {p}_{j}=1| {\beta }_{j}| .]\]\end{document}
Establishment of the SVM model
The SVM model was constructed by integrating clinical factors with CT image features. A non-linear mapping function (Kernel function) was employed to transform the data into a high-dimensional feature space, enhancing the potential linear separability of data within this space. Subsequently, an optimal hyperplane was determined within the feature space, effectively delineating distinct categories of samples. Support vectors, representing the samples closest to the hyperplane, were utilized to compute the distance between support vectors and the hyperplane. This computation facilitated the establishment of decision boundaries and the basis for discriminating new samples. The effectiveness of the models for both the research and control groups was validated. The model is illustrated in Figure 2 .
Diagnostic criteria
According to the Diagnostic Criteria and Therapeutic Principle of Invasive Fungal Infection in Hematological Diseases or Malignant Tumors [ 26 ], the diagnosis of PFI was made. The diagnosis was divided into confirmed, clinically confirmed, probable, and undetermined. Histopathological examination was conducted with the specimens taken from patients included, from which molds were shown with hyphae, yeasts showed yeast cells or hyphae, and sporozoites showed sporozoite cysts or trophozoites.
Observation indicators
The basic data (age, gender, and length of hospital stay) of the patients were collected. The disease types of HM were counted, including acute lymphoblastic leukemia (ALL), myelodysplastic syndrome (MDS), non-Hodgkin’s lymphoma (NHL), and multiple myeloma (MM). The strain categories and their proportions ( Aspergillus , Mucor , Candida albicans , Cryptococcus, and Pneumocystis carinii ) were also counted. The CT image features of the patients were statistically analyzed based on the type of fungus infecting the patients. These features included lesion characteristics, diffuse lung lesions, lesion extent, lesion morphology, air crescent sign, and nodule size. Lesion characteristics encompassed multiple lesions and single lesions, diffuse lung lesions included unilateral lung lesions and bilateral lung lesions, and lesion extent covered subpleural lesions, perihilar pulmonary lesions, and other lung areas. Lesion morphology included single form and single form. The air crescent sign represented the formation of a crescent-shaped lucency between spherical lesions or cavities within the lungs and the cavity wall. Nodule size was categorized into patients with nodules measuring ≤3 and >3 cm.
Statistical methods
SPSS19.0 was applied for data processing in this study. The measurement data and the enumeration data were expressed as mean ± standard deviation ( x ¯ \documentclass[10pt]{article}\usepackage{wasysym}\usepackage[substack]{amsmath}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage[mathscr]{eucal}\usepackage{mathrsfs}\usepackage{pmc}\usepackage[Euler]{upgreek}\pagestyle{empty}\oddsidemargin -1.0in\begin{document}\[\bar{x}]\]\end{document} ± s ) and percentage (%), respectively. Pairwise comparisons were made using one-way analysis of variance. The SVM model was employed for diagnostic purposes and assessed the diagnostic efficiency of the model using the receiver operating characteristic (ROC) curve and the AUC. The difference was considered statistically significance at P < 0.05.
Inclusion criteria: The patients had complete clinical data, normal intelligence, and barrier-free communication, as well as a length of hospital stay for more than 48 h.
Exclusion criteria: Patients with psychiatric diseases, those who dropped out of the research due to personal reasons, those who had poor CT image quality, and those with difficulty in listening, writing, and normal communication were excluded.
Informed consent: Informed consent has been obtained from all individuals included in this study.
Ethical approval: The research related to human use has complied with all the relevant national regulations, and institutional policies in accordance with the tenets of the Helsinki Declaration, and has been approved by the ethics committee of the hospital.
Results
Basic data of patients with HM
Among the 100 patients in the training set with malignant hematological diseases complicated by secondary PFI, 68 were male, and 32 were female. Sixty-six patients were ≥55 years old and 34 patients were less than 55 years old. There were 41 patients with hospital stay ≥30 days, while 59 patients with ≤30 days. For the disease types, 38, 26, 23, and 13 cases had ALL, MDS, NHL, and MM, respectively ( Figure 3 ).
In the test set of 80 patients with malignant hematological diseases, there were 51 males and 29 females. Among these, 48 patients were aged 55 years or older, while 32 were younger than 55 years. Additionally, there were 33 patients with hospital stays of 30 days or more and 47 patients with stays of less than 30 days. Regarding disease types, there were 29 patients with ALL, 19 with MDS, 19 with NHL, and 13 with MM ( Figure 4 ).
The analysis of data pertaining to patient characteristics, including gender, age, and disease type, in both the training set and the test set revealed no significant statistical differences ( P > 0.05).
Strain categories in patients
As shown in Figure 5 , among the 100 patients with HM complicated by secondary PFI, the most common pathogen was Aspergillus found in 40 cases (40%). It was followed by Candida albicans in 28 cases (28%) and then Pneumocystis carinii found in 16 cases (16%). The fewest pathogens were Mucor and Cryptococcus , which were found in 9 cases (9%) and 7 cases (7%), respectively.
CT features of different PFIs
As shown in Figure 6 , among patients with fungal infections such as Aspergillus, Mucor, Candida, Cryptococcus, and Pneumocystis , when observing the lesion characteristics, multiple lesions were most common in patients with Aspergillus infections. Regarding lesion characteristics, patients with diffuse lung lesions had a significantly higher incidence of bilateral lung infections compared to those with unilateral lung infections. Among these, patients with bilateral lung infections in Aspergillus were the most common, followed by Candida, Pneumocystis, Mucor , and Cryptococcus . Furthermore, patients with Aspergillus, Mucor, and Cryptococcus infections mainly had lesions below the pleura, while those with Candida and Pneumocystis infections predominantly had lesions in other lung regions. In terms of lesion morphology, patients with polymorphic infections were significantly more common than those with a single form among those with Aspergillus, Mucor, Candida, Cryptococcus, and Pneumocystis fungal infections. When observing patients with air crescent sign, irregular cavities, pleural adhesions, it was found that patients with Aspergillus infections had the highest number of patients with air crescent sign, followed by Pneumocystis, Cryptococcus, Candida, and Mucor infections. Looking at the size of nodules in different infection patients, it was found that patients with Aspergillus, Mucor, and Cryptococcus infections had significantly more patients with nodules smaller than 3 cm than those with nodules larger than or equal to 3 cm. On the other hand, patients with Mucor and Pneumocystis infections had significantly fewer patients with nodules smaller than 3 cm compared to those with nodules larger than or equal to 3 cm.
Extraction of image features
A total of 24 features were selected from the CT images, and after processing, the six most influential radiomics features were extracted. These features are Squarerooted Grey-Level Co-occurrence Matrix Inverse Variance (Squareroot-IV), Squared Grey-Level Co-occurrence Matrix Inverse Variance (Square-IV), Exponential Maximum Correlation Coefficient (E-MCC), Square-rooted Maximum Correlation Coefficient (s-MCC), First-order Kurtosis (FK), and Small Dependence Low Grey-Level Emphasis (SDLGLE). As shown in Figure 7 , the relative importance ranking of these six radiomics features from highest to lowest was Squareroot-IV, FK, E-MCC, Square-IV, SDLGLE, and s-MCC.
Prediction efficiency of the SVM model
The diagnostic model consisted of clinical data for malignant hematological disorders complicated by secondary PFI. CT images showed ground-glass opacities in the upper lobe of the left lung, with surrounding halo signs; nodular shadows within the lungs were enlarged and exhibited consolidation, with nodules along the walls appearing in the left upper lobe; diffuse small patchy shadows were observed in the right lung, accompanied by irregularly shaped air crescent signs in the periphery. Six radiomics features were extracted ( Figure 8 ). In the training set, the AUC of the SVM model was 0.902 (95% CI: 0.803–0.924). In the testing set, the model achieved an AUC of 0.891 (95% CI: 0.793–0.901).
Imaging data
Figure 9 displays the CT data of a male patient, and the sputum fungal culture of Aspergillus flavus was made. Chest CT ( Figure 9a and b ) showed a patchy ground-glass opacity in the upper lobe of the left lung with a surrounding halo sign. First, the patient was given anti-infective treatment such as cefoperazone, imipenem, and vancomycin combined with voriconazole, but the fever symptoms did not improve obviously. CT examination ( Figure 9c and d ) was performed again, from which the nodules in the lungs were larger than before, accompanied by consolidation, and wall nodules appeared in the left upper lobe. The patient was then treated with posaconazole, glucocorticoid, and other treatments. After 2 months, the CT re-examination ( Figure 9e and f ) showed that the left-lung infiltration shadow disappeared, but there were residual linear and patch shadows.
Figure 10 presents the CT data of another male patient, and the washing fluid culture was conducted with Candida albicans . Chest CT ( Figure 10a and b ) revealed diffuse small patchy shadows in the right lung with an air meniscus sign and irregular borders. First, the patient was treated with voriconazole and caspofungin. After 1 month, an examination was given. The patient was still coughing, and CT ( Figure 10c and d ) showed that the changes were not notable in the consolidation shadow of the hilum of the right lung, and the patch shadows were not reduced much. After that, amphotericin B and glucocorticoid therapy were given. The CT scan after 2 months ( Figure 10e and f ) suggested that the consolidation shadow of the right hilum disappeared. The patchy shadows in the lungs shrank, and cavities appeared in the patch in the lateral segment of the right middle lobe.
Prognosis of patients with different PFIs
Figure 11 displays the prognosis of patients with different PFIs. Among patients with Aspergillus -induced PFI, 10 cases (25%) recovered, 21 cases (52.5%) had chronic or organized lesions, 4 cases (10%) suffered from the recurrent disease without recovery, and 5 cases (12.5%) died. Among patients with PFI induced by Mucor , 1 case (11.11%), 2 cases (22.22%), 1 case (11.11%), and 1 case (11.11%) recovered had chronic or organized lesions, unrecovered with recurrent disease, and died, respectively. Among patients with PFI of Candida albicans , 7 (25%), 15 (53.57%), 4 (14.29%), and 2 (7.14%) cases had the above consequences, respectively. Among patients with Cryptococcus -induced PFI, 0 cases (0%), 4 cases (57.14%), 2 cases (28.57%), and 1 case (14.29%) encountered the above, respectively. Among patients with PFI of Pneumocystis carinii , 7 cases (43.75%) recovered,3 cases (18.75%) had chronic or organized lesions, 2 cases (12.5%) had the disease recurrently, and 4 cases (25%) died. Patients with Mucor -induced PFI had the highest mortality, followed by Pneumocystis carinii . Patients with PFI of Pneumocystis carinii had the highest recovery rate.
Discussion
Pulmonary infection is the most common symptom of infection, accounting for about 40% of all infections. With the improvement of infection diagnosis and treatment capabilities and the use of new antibiotics in recent years, great progress has been made in the control and management of pulmonary infection in clinical practice. Most patients including those with bacterial infections and fungal infections can be cured [ 27 – 29 ]. Fungal infections have a high incidence in patients with hematological diseases, and some patients have poor prognosis, which affects the course of treatment, treatment costs, and mortality of the underlying diseases to a certain extent. Therefore, it is very necessary to understand the features of different fungal infections in patients with HM [ 30 ]. In this research, the results indicated that Aspergillus and Candida albicans were the major pathogenic bacteria in HM complicated by PFI [ 31 ].
CT imaging has significant clinical utility in patients with secondary PFI complicating HM. Due to their compromised immune function, patients with HM are susceptible to bacterial, viral, and fungal infections, with PFIs being a common and severe complication [ 29 , 32 , 33 ]. CT imaging provides detailed information about pulmonary structures, aiding in the early detection and assessment of the extent of PFI, thereby guiding treatment strategies. In patients with secondary PFI complicating HM, CT imaging provides following valuable insights: (1) Location and distribution of pulmonary lesions: CT can clearly depict the location and distribution of pulmonary infection foci, including unilateral or bilateral involvement, and localized or diffuse distribution, assisting in the determination of the extent and severity of the infection. (2) Morphology and density of lesions: CT can delineate the morphological features of infection foci, such as nodules, cavities, and consolidation, and evaluate lesion density. Different pathogens may cause infections with distinct morphological and density characteristics; thus, CT imaging can offer clues regarding the type of pathogen involved. (3) Signs of concurrent complications: patients with secondary PFI complicating HM often present with complications such as pleural effusion, air bronchogram signs, and pleural effusion. CT aids in detecting the presence and assessing the severity of these complications, providing a basis for a comprehensive evaluation of the patient’s condition. Studies have demonstrated that CT imaging is highly sensitive for the early diagnosis of invasive pulmonary aspergillosis complicating HM. Furthermore, Longhitano et al. [ 34 ] have shown that CT imaging can reflect the presence of disseminated infection and is highly sensitive for monitoring infection progression over time. In this research, the CT characteristics of different PFI patients were analyzed, and the results revealed that multiple lesions are most commonly observed in patients with Aspergillus PFI. The diffuse lung lesions were observed in bilateral lungs, and the lesions were mainly subpleural. Air meniscus signs, irregular cavities, and pleural adhesions could be observed, and the nodules were basically less than 3 cm. Multiple lesions were also the most common in patients with Mucor -induced PFI. The diffuse lung lesions in bilateral lungs and the mainly subpleural lesions were on display. None of the air meniscus signs was discovered, and the nodules were mostly less than 3 cm. Patients with Cryptococcus-induced PFI were also predominant with multiple lesions, together with bilateral-lung diffuse lesions, no air meniscus sign, irregular cavities, and nodules less than 3 cm. All patients with PFI of Candida albicans had multiple lesions, with various morphologies, bilateral-lung diffuse lesions, most lesions in other lung regions, and nodules more than 3 cm. Thus, from the CT image features, it could be concluded that patients with different PFIs were shown differently in terms of lesion condition, lesion extent, lesion morphology, and nodule size. Therefore, CT could be utilized for the assessment of PFIs in patients with hematological diseases. From the prognosis of the patients, the patients with Mucor -PFI had the highest mortality (55.56%), followed by 25% of Pneumocystis carinii -induced PFI. The patients with Pneumocystis carinii -PFI had the highest recovery rate of 43.75%. Ghuman et al. [ 35 ] suggested that mucormycosis affecting the intestinal tract had a high mortality, and early identification and intervention could considerably improve the prognosis of patients. The disease should be suspected in immunosuppressed patients as their imaging findings showed unexplained bowel ischemia, infarction, and/or pneumonia without any obvious vascular thrombosis. The results here were similar.
Analysis revealed no significant statistical differences in gender, age, and disease type between the training and testing sets. This suggests a relatively balanced distribution of patient gender, age, and disease type in both the training and testing datasets. The lack of significant statistical differences indicates a similarity in data distribution between the training and testing sets, enhancing the model’s generalization ability in real-world scenarios. In terms of image feature extraction, six optimal radiomics features were extracted from the 24 CT images. These included Squareroot-IV, Square-IV, E-MCC, s-MCC, FK, and SDLGLE. These features are considered to play a significant role in the diagnostic model for malignant hematological disorders complicated by secondary PFI. Squareroot-IV, for example, can be utilized to characterize the complexity of texture in images, while FK reflects peakness information in the grayscale distribution of images [ 36 ]. The extraction of these features aids the model in comprehending latent diagnostic information within the images, thereby enhancing predictive accuracy. The evaluation of the SVM model’s predictive performance on the training and testing sets involves measuring its performance through the calculation of the AUC of the ROC curve. The SVM model yielded an AUC of 0.902 (95% CI: 0.803–0.924) on the training set, and an AUC of 0.891 (95% CI: 0.793–0.901) on the testing set. A higher AUC value closer to 1 indicates better model performance. This suggests that the SVM model demonstrates strong predictive capability in diagnosing malignant hematological disorders complicated by secondary PFI. The results indicate that the model can accurately predict whether a patient is affected by PFI based on their CT images and radiomics features, thereby providing a basis for assisting medical professionals in making informed decisions.
Conclusion
In this research, CT examinations were conducted on a cohort of 100 patients with HM complicated by secondary PFI. The findings revealed that the nature of lesions, their extent, morphology, and nodule sizes varied among patients with different PFIs. Therefore, CT imaging can be effectively utilized for assessing PFIs in hematological disease patients. Furthermore, the SVM model constructed based on extracted radiomic features exhibited strong predictive performance in diagnosing secondary PFIs in patients with HM. These results indicate that SVM algorithm models can be employed for rapid diagnosis when dealing with the diagnosis of secondary PFIs in HM. These findings hold significant clinical relevance for improving the diagnosis of complications in HM. This research had a relatively small sample size, consisting of only 100 patients with HM complicated by secondary PFI. Future research could benefit from expanding the sample size, including a more diverse range of hematological malignancy types, to validate and generalize the applicability of this diagnostic approach. While the SVM algorithm model demonstrated strong predictive capabilities in diagnosing secondary PFI, there is scope for further optimization and improvement of the model performance. Future research efforts may explore the use of other machine learning algorithms or deep learning methods to enhance diagnostic accuracy and stability. Additionally, future studies could continue to investigate the influence of different pathogen types on CT imaging features and develop more precise prognostic assessment models.
Funding information: Authors state no funding involved.
Conflict of interest: Authors state no conflict of interest.
Data availability statement : The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
References
[1] Gaudio A, Xourafa A, Rapisarda R, Zanoli L, Signorelli SS, Castellino P. Hematological diseases and osteoporosis. Int J Mol Sci. 2020;21(10):3538. 10.3390/ijms21103538 PMC7279036 32429497 Gaudio A Xourafa A Rapisarda R Zanoli L Signorelli SS Castellino P Hematological diseases and osteoporosis Int J Mol Sci 2020 21 10 3538 32429497 10.3390/ijms21103538 PMC7279036
[2] Klein A, Molad Y. Hematological manifestations among patients with rheumatic diseases. Acta Haematol. 2021;144(4):403–12. 10.1159/000511759 33221805 Klein A Molad Y Hematological manifestations among patients with rheumatic diseases Acta Haematol 2021 144 4 403 12 33221805 10.1159/000511759
[3] Liao XY, Qiu KY, Wu RH, Guo SY, Wang J, Huang K, et al. Clinical analysis of 164 children of blood disease complicated with invasive fungal disease. Zhongguo Shi Yan Xue Ye Xue Za Zhi. 2019;27(5):1672–7 (Chinese). 10.19746/j.cnki.issn.1009-2137.2019.05.048 31607330 Liao XY Qiu KY Wu RH Guo SY Wang J Huang K Clinical analysis of 164 children of blood disease complicated with invasive fungal disease Zhongguo Shi Yan Xue Ye Xue Za Zhi 2019 27 5 1672 7 (Chinese). 31607330 10.19746/j.cnki.issn.1009-2137.2019.05.048
[4] Kohn DB. Gene therapy for blood diseases. Curr Opin Biotechnol. 2019;60:39–45. 10.1016/j.copbio.2018.11.016 30599357 Kohn DB Gene therapy for blood diseases Curr Opin Biotechnol 2019 60 39 45 30599357 10.1016/j.copbio.2018.11.016
[5] Wagner SJ, Leiby DA, Roback JD. Existing and emerging blood-borne pathogens: Impact on the safety of blood transfusion for the hematology/oncology patient. Hematol Oncol Clin North Am. 2019;33(5):739–48. 10.1016/j.hoc.2019.05.002 31466601 Wagner SJ Leiby DA Roback JD Existing and emerging blood-borne pathogens: Impact on the safety of blood transfusion for the hematology/oncology patient Hematol Oncol Clin North Am 2019 33 5 739 48 31466601 10.1016/j.hoc.2019.05.002
[6] Barriteau CM, Bochey P, Lindholm PF, Hartman K, Sumugod R, Ramsey G. Blood transfusion utilization in hospitalized COVID-19 patients. Transfusion. 2020;60(9):1919–23. 10.1111/trf.15947 PMC7361376 32583506 Barriteau CM Bochey P Lindholm PF Hartman K Sumugod R Ramsey G Blood transfusion utilization in hospitalized COVID-19 patients Transfusion 2020 60 9 1919 23 32583506 10.1111/trf.15947 PMC7361376
[7] Krishnamurti L, Ross D, Sinha C, Leong T, Bakshi N, Mittal N, et al. Comparative effectiveness of a web-based patient decision aid for therapeutic options for sickle cell disease: Randomized controlled trial. J Med Internet Res. 2019;21(12):e14462. 10.2196/14462 PMC6934048 31799940 Krishnamurti L Ross D Sinha C Leong T Bakshi N Mittal N Comparative effectiveness of a web-based patient decision aid for therapeutic options for sickle cell disease: Randomized controlled trial J Med Internet Res 2019 21 12 e14462 31799940 10.2196/14462 PMC6934048
[8] Hughes S, Troise O, Donaldson H, Mughal N, Moore LSP. Bacterial and fungal coinfection among hospitalized patients with COVID-19: A retrospective cohort study in a UK secondary-care setting. Clin Microbiol Infect. 2020;26(10):1395–9. 10.1016/j.cmi.2020.06.025 PMC7320692 32603803 Hughes S Troise O Donaldson H Mughal N Moore LSP Bacterial and fungal coinfection among hospitalized patients with COVID-19: A retrospective cohort study in a UK secondary-care setting Clin Microbiol Infect 2020 26 10 1395 9 32603803 10.1016/j.cmi.2020.06.025 PMC7320692
[9] Garcia-Vidal C, Sanjuan G, Moreno-García E, Puerta-Alcalde P, Garcia-Pouton N, Chumbita M, et al. COVID-19 Researchers Group. Incidence of co-infections and superinfections in hospitalized patients with COVID-19: A retrospective cohort study. Clin Microbiol Infect. 2021;27(1):83–8. 10.1016/j.cmi.2020.07.041 PMC7836762 32745596 Garcia-Vidal C Sanjuan G Moreno-García E Puerta-Alcalde P Garcia-Pouton N Chumbita M COVID-19 Researchers Group. Incidence of co-infections and superinfections in hospitalized patients with COVID-19: A retrospective cohort study Clin Microbiol Infect 2021 27 1 83 8 32745596 10.1016/j.cmi.2020.07.041 PMC7836762
[10] Durand ML. Bacterial and fungal endophthalmitis. Clin Microbiol Rev. 2017;30(3):597–613. 10.1128/CMR.00113-16 PMC5475221 28356323 Durand ML Bacterial and fungal endophthalmitis Clin Microbiol Rev 2017 30 3 597 613 28356323 10.1128/CMR.00113-16 PMC5475221
[11] Jenks JD, Cornely OA, Chen SC, Thompson GR 3rd, Hoenigl M. Breakthrough invasive fungal infections: Who is at risk? Mycoses. 2020;63(10):1021–32. 10.1111/myc.13148 32744334 Jenks JD Cornely OA Chen SC Thompson GR Hoenigl M Breakthrough invasive fungal infections: Who is at risk? Mycoses 2020 63 10 1021 32 32744334 10.1111/myc.13148
[12] Bayram N, Ozsaygılı C, Sav H, Tekin Y, Gundogan M, Pangal E, et al. Susceptibility of severe COVID-19 patients to rhino-orbital mucormycosis fungal infection in different clinical manifestations. Jpn J Ophthalmol. 2021;65(4):515–25. 10.1007/s10384-021-00845-5 PMC8165350 34057620 Bayram N Ozsaygılı C Sav H Tekin Y Gundogan M Pangal E Susceptibility of severe COVID-19 patients to rhino-orbital mucormycosis fungal infection in different clinical manifestations Jpn J Ophthalmol 2021 65 4 515 25 34057620 10.1007/s10384-021-00845-5 PMC8165350
[13] Leroy-Freschini B, Treglia G, Argemi X, Bund C, Kessler R, Herbrecht R, et al. 18F-FDG PET/CT for invasive fungal infection in immunocompromised patients. QJM. 2018;111(9):613–22. 10.1093/qjmed/hcy128 29917146 Leroy-Freschini B Treglia G Argemi X Bund C Kessler R Herbrecht R 18F-FDG PET/CT for invasive fungal infection in immunocompromised patients QJM 2018 111 9 613 22 29917146 10.1093/qjmed/hcy128
[14] Wongchai A, Jenjeti D, Priyadarsini AI, Deb N, Bhardwaj A, Tomar P. Farm monitoring and disease prediction by classification based on deep learning architectures in sustainable agriculture. Ecol Model. 2022;474:110167. Wongchai A Jenjeti D Priyadarsini AI Deb N Bhardwaj A Tomar P Farm monitoring and disease prediction by classification based on deep learning architectures in sustainable agriculture Ecol Model 2022 474 110167
[15] Malik I, Bhardwaj A, Bhardwaj H, Sakalle A. IoT-Enabled Smart Homes: Architecture, Challenges, and Issues. In: Mishra D, Sharma S, editors. Revolutionizing Industrial Automation Through the Convergence of Artificial Intelligence and the Internet of Things (Vol. 2023). IGI Global; 2023. p. 160–76. Malik I Bhardwaj A Bhardwaj H Sakalle A IoT-Enabled Smart Homes: Architecture, Challenges, and Issues Mishra D Sharma S Revolutionizing Industrial Automation Through the Convergence of Artificial Intelligence and the Internet of Things Vol. 2023 IGI Global 2023 p. 160 76
[16] Kumar A, Sinha N, Bhardwaj A, Goel S. Clinical risk assessment of chronic kidney disease patients using genetic programming. Comput Methods Biomech Biomed Eng. 2022;25(8):887–95. 10.1080/10255842.2021.1985476 34726985 Kumar A Sinha N Bhardwaj A Goel S Clinical risk assessment of chronic kidney disease patients using genetic programming Comput Methods Biomech Biomed Eng 2022 25 8 887 95 10.1080/10255842.2021.1985476 34726985
[17] Shafi I, Din S, Khan A, Díez IT, Casanova RDJP, Pifarre KT, et al. An effective method for lung cancer diagnosis from CT scan using deep learning-based support vector network. Cancers (Basel). 2022;14(21):5457. 10.3390/cancers14215457 PMC9657078 36358875 Shafi I Din S Khan A Díez IT Casanova RDJP Pifarre KT An effective method for lung cancer diagnosis from CT scan using deep learning-based support vector network Cancers (Basel) 2022 14 21 5457 36358875 10.3390/cancers14215457 PMC9657078
[18] Turukmane AV, Alhebaishi N, Alshareef AM, Mirza OM, Bhardwaj A, Singh B. Multispectral image analysis for monitoring by IoT based wireless communication using secure locations protocol and classification by deep learning techniques. Optik. 2022;271:170122. Turukmane AV Alhebaishi N Alshareef AM Mirza OM Bhardwaj A Singh B Multispectral image analysis for monitoring by IoT based wireless communication using secure locations protocol and classification by deep learning techniques Optik 2022 271 170122
[19] Yan C, Hao P, Wu G, Lin J, Xu J, Zhang T, et al. Machine learning-based combined nomogram for predicting the risk of pulmonary invasive fungal infection in severely immunocompromised patients. Ann Transl Med. 2022;10(9):514. 10.21037/atm-21-4980 PMC9347049 35928747 Yan C Hao P Wu G Lin J Xu J Zhang T Machine learning-based combined nomogram for predicting the risk of pulmonary invasive fungal infection in severely immunocompromised patients Ann Transl Med 2022 10 9 514 35928747 10.21037/atm-21-4980 PMC9347049
[20] Godoy MCB, Ferreira Dalla Pria HR, Truong MT, Shroff GS, Marom EM. Invasive fungal pneumonia in immunocompromised patients. Radiol Clin North Am. 2022;60(3):497–506. 10.1016/j.rcl.2022.01.006 35534133 Godoy MCB Ferreira Dalla Pria HR Truong MT Shroff GS Marom EM Invasive fungal pneumonia in immunocompromised patients Radiol Clin North Am 2022 60 3 497 506 35534133 10.1016/j.rcl.2022.01.006
[21] Lewis RE, Stanzani M, Morana G, Sassi C. Radiology-based diagnosis of fungal pulmonary infections in high-risk hematology patients: are we making progress? Curr Opin Infect Dis. 2023;36(4):250–6. 10.1097/QCO.0000000000000937 PMC10351900 37431554 Lewis RE Stanzani M Morana G Sassi C Radiology-based diagnosis of fungal pulmonary infections in high-risk hematology patients: are we making progress? Curr Opin Infect Dis 2023 36 4 250 6 37431554 10.1097/QCO.0000000000000937 PMC10351900
[22] Kunihiro Y, Tanaka N, Kawano R, Yujiri T, Ueda K, Gondo T, et al. High-resolution CT findings of pulmonary infections in patients with hematologic malignancy: comparison between patients with or without hematopoietic stem cell transplantation. Jpn J Radiol. 2022;40(8):791–9. 10.1007/s11604-022-01260-7 PMC9345826 35284995 Kunihiro Y Tanaka N Kawano R Yujiri T Ueda K Gondo T High-resolution CT findings of pulmonary infections in patients with hematologic malignancy: comparison between patients with or without hematopoietic stem cell transplantation Jpn J Radiol 2022 40 8 791 9 35284995 10.1007/s11604-022-01260-7 PMC9345826
[23] Hałaburda-Rola M, Dzieciątkowski T, Górka M, Rowiński O, Grabowska-Derlatka L. Clinical utility of the updated European Organization for Research and Treatment of Cancer/Invasive Fungal Infections Cooperative Group and the National Institute of Allergy and the Mycoses Study Group Education and Research Consortium computed tomography criteria of invasive pulmonary aspergillosis in hematological malignancies. Hematology. 2021;26(1):398–407. 10.1080/16078454.2021.1931739 34057050 Hałaburda-Rola M Dzieciątkowski T Górka M Rowiński O Grabowska-Derlatka L Clinical utility of the updated European Organization for Research and Treatment of Cancer/Invasive Fungal Infections Cooperative Group and the National Institute of Allergy and the Mycoses Study Group Education and Research Consortium computed tomography criteria of invasive pulmonary aspergillosis in hematological malignancies Hematology 2021 26 1 398 407 34057050 10.1080/16078454.2021.1931739
[24] Martinez D, Ananda-Rajah MR, Suominen H, Slavin MA, Thursky KA, Cavedon L. Automatic detection of patients with invasive fungal disease from free-text computed tomography (CT) scans. J Biomed Inf. 2015;53:251–60. 10.1016/j.jbi.2014.11.009 25460203 Martinez D Ananda-Rajah MR Suominen H Slavin MA Thursky KA Cavedon L Automatic detection of patients with invasive fungal disease from free-text computed tomography (CT) scans J Biomed Inf 2015 53 251 60 10.1016/j.jbi.2014.11.009 25460203
[25] Qian X, Rong H, Wei X, Rong G, Yao M. Value of CT radiomics combined with clinical features in the diagnosis of allergic bronchopulmonary aspergillosis. Comput Math Methods Med. 2022;2022:5317509. 10.1155/2022/5317509 PMC9098310 35572830 Qian X Rong H Wei X Rong G Yao M Value of CT radiomics combined with clinical features in the diagnosis of allergic bronchopulmonary aspergillosis Comput Math Methods Med 2022 2022 5317509 35572830 10.1155/2022/5317509 PMC9098310
[26] Shen ZX. Diagnostic criteria and therapeutic principle of invasive fungal infection in hematological diseases or malignant tumors. Zhonghua Nei Ke Za Zhi. 2007 Jul;46(7):532–3 (Chinese). 17959068 Shen ZX Diagnostic criteria and therapeutic principle of invasive fungal infection in hematological diseases or malignant tumors Zhonghua Nei Ke Za Zhi 2007 Jul 46 7 532 3 (Chinese). 17959068
[27] Sanguinetti M, Posteraro B, Beigelman-Aubry C, Lamoth F, Dunet V, Slavin M, et al. Diagnosis and treatment of invasive fungal infections: Looking ahead. J Antimicrob Chemother. 2019;74(Suppl 2):ii27–37. 10.1093/jac/dkz041 31222314 Sanguinetti M Posteraro B Beigelman-Aubry C Lamoth F Dunet V Slavin M Diagnosis and treatment of invasive fungal infections: Looking ahead J Antimicrob Chemother 2019 74 Suppl 2 ii27 37 31222314 10.1093/jac/dkz041
[28] Yan C, Xu J, Liang C, Wei Q, Wu Y, Xiong W, et al. Radiation dose reduction by using CT with iterative model reconstruction in patients with pulmonary invasive fungal infection. Radiology. 2018;288(1):285–92. 10.1148/radiol.2018172107 29634436 Yan C Xu J Liang C Wei Q Wu Y Xiong W Radiation dose reduction by using CT with iterative model reconstruction in patients with pulmonary invasive fungal infection Radiology 2018 288 1 285 92 29634436 10.1148/radiol.2018172107
[29] Wang JW, Yang FF, Zhang CY, Lin JZ, Wang HX, Xu WJ. Imaging characteristics of invasive pulmonary fungal infection secondary to hematological diseases and comparison before and after treatment. J Healthc Eng. 2021;2021:3736108. 10.1155/2021/3736108 PMC8500768 34630984 Wang JW Yang FF Zhang CY Lin JZ Wang HX Xu WJ Imaging characteristics of invasive pulmonary fungal infection secondary to hematological diseases and comparison before and after treatment J Healthc Eng 2021 2021 3736108 34630984 10.1155/2021/3736108 PMC8500768
[30] Angelini E, Shah A. Using artificial intelligence in fungal lung disease: CPA CT imaging as an example. Mycopathologia. 2021;186(5):733–7. 10.1007/s11046-021-00546-0 PMC8536566 33840005 Angelini E Shah A Using artificial intelligence in fungal lung disease: CPA CT imaging as an example Mycopathologia 2021 186 5 733 7 33840005 10.1007/s11046-021-00546-0 PMC8536566
[31] Douglas AP, Thursky KA, Worth LJ, Harrison SJ, Hicks RJ, Slavin MA. FDG-PET/CT in managing infection in patients with hematological malignancy: Clinician knowledge and experience in Australia. Leuk Lymphoma. 2019;60(10):2471–6. 10.1080/10428194.2019.1590571 30947578 Douglas AP Thursky KA Worth LJ Harrison SJ Hicks RJ Slavin MA FDG-PET/CT in managing infection in patients with hematological malignancy: Clinician knowledge and experience in Australia Leuk Lymphoma 2019 60 10 2471 6 30947578 10.1080/10428194.2019.1590571
[32] Seldeslachts L, Jacobs C, Tielemans B, Vanhoffelen E, Van der Sloten L, Humblet-Baron S, et al. Overcome double trouble: Baloxavir marboxil suppresses influenza thereby mitigating secondary invasive pulmonary Aspergillosis. J Fungi (Basel). 2021;8(1):1. 10.3390/jof8010001 PMC8777735 35049941 Seldeslachts L Jacobs C Tielemans B Vanhoffelen E Van der Sloten L Humblet-Baron S Overcome double trouble: Baloxavir marboxil suppresses influenza thereby mitigating secondary invasive pulmonary Aspergillosis J Fungi (Basel) 2021 8 1 1 35049941 10.3390/jof8010001 PMC8777735
[33] Sharma R, Singh C, Khadwal A, Prakash G, Malhotra P, Jain A, et al. Role of pre-transplant chest high-resolution computed tomography and serum galactomannan index in predicting post-transplant invasive pulmonary aspergillosis in allogeneic hematopoietic cell transplant recipients. Transpl Infect Dis. 2021;23(4):e13632. 10.1111/tid.13632 33973316 Sharma R Singh C Khadwal A Prakash G Malhotra P Jain A Role of pre-transplant chest high-resolution computed tomography and serum galactomannan index in predicting post-transplant invasive pulmonary aspergillosis in allogeneic hematopoietic cell transplant recipients Transpl Infect Dis 2021 23 4 e13632 33973316 10.1111/tid.13632
[34] Longhitano A, Alipour R, Khot A, Bajel A, Antippa P, Slavin M, et al. The role of 18F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography (FDG PET/CT) in assessment of complex invasive fungal disease and opportunistic co-infections in patients with acute leukemia prior to allogeneic hematopoietic cell transplant. Transpl Infect Dis. 2021;23(3):e13547. 10.1111/tid.13547 33338319 Longhitano A Alipour R Khot A Bajel A Antippa P Slavin M The role of 18F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography (FDG PET/CT) in assessment of complex invasive fungal disease and opportunistic co-infections in patients with acute leukemia prior to allogeneic hematopoietic cell transplant Transpl Infect Dis 2021 23 3 e13547 33338319 10.1111/tid.13547
[35] Ghuman SS, Sindhu P, Buxi TBS, Sheth S, Yadav A, Rawat KS, et al. CT appearance of gastrointestinal tract mucormycosis. Abdom Radiol (NY). 2021;46(5):1837–45. 10.1007/s00261-020-02854-3 33170347 Ghuman SS Sindhu P Buxi TBS Sheth S Yadav A Rawat KS CT appearance of gastrointestinal tract mucormycosis Abdom Radiol (NY) 2021 46 5 1837 45 33170347 10.1007/s00261-020-02854-3
[36] Huang T, Zheng X, He L, Chen Z. Diagnostic value of deep learning-based CT feature for severe pulmonary infection. J Healthc Eng. 2021;2021:5359084. 10.1155/2021/5359084 PMC8641994 34868521 Huang T Zheng X He L Chen Z Diagnostic value of deep learning-based CT feature for severe pulmonary infection J Healthc Eng 2021 2021 5359084 34868521 10.1155/2021/5359084 PMC8641994