Lumbar Spine Disease Detection: Enhanced CNN Model With Improved Classification Accuracy

Back pain is an issue affecting millions of people throughout the world. Research on back pain root cause detection is immense. The lumbar spine is a lower back region of the backbone that could also be responsible for back pain. Research on issues related to lumbar spine disease (LSD) is limited. The lumbar spine is critical for the human body as it supports the weight of the body. Many diseases can affect the lumbar spine adversely. This research is directed towards the detection of LSDs using optimized feature extraction and selection phases. Furthermore, the linearity-based model is used for feature selection, selecting only the best possible features to reduce the missed classification degree. The flow of the proposed research work is divided into phases. In the first phase, data collection is performed. Data collection in the proposed work includes both real-time and benchmark magnetic resonance imaging (MRI) datasets. The MRI dataset collected from the hospital is validated by medical experts. After data collection, pre-processing is applied. This step removes the noise from the collected dataset. The pre-processing mechanism is performed using histogram equalization, median filtering, validation, and normalization. After the pre-processing phase, background subtraction and region of interest (ROI) detection are performed using the region-cut mechanism. Optimal feature extraction is achieved using a differential spider monkey optimization (SMO), and feature selection is performed using a linearity-based convolutional neural network (CNN) model. In the end, ensemble-based classification is used for disease prediction. The validation of the result is conducted through classification accuracy, specificity, sensitivity, and F-Score. The high classification accuracy of 96% is achieved with multi support vector machine (MSVM), 94% with random forest (RF), 93.5% with a decision tree (DT), and 91% with the Naïve Bayes (NB) approach, proving the validity of the proposed approach.


I. INTRODUCTION
The lumbar spine is at the lower end of the spinal cord and is required for balancing the weight of the human body, The associate editor coordinating the review of this manuscript and approving it for publication was Marco Giannelli .as discussed in [1].The spinal cord area is shielded by five durable and flexible vertebrae that ensure the dispersion of axial forces, as discussed in [2].The spinal cord goes through the vertebrae and terminates at the level of the L1 and L2 vertebrae.The cauda equina begins at the termination of the spinal cord and descends through the remainder of the canal,  as discussed in [3].The lumbar spine is made up of bones, cartilage, nerves, and muscles.Each of these components plays a critical role in the formation and operation of the lumbar spine [4].The vital functions of the lumbar spine are shown in Figure 1.
There are three different vital functions associated with the lumbar spine.The primary one is the protection of the spinal cord and nerves [5].Besides, it bears the upper body weight and provides support to the neck, head, and trunk [1].The next one is truncal motion, including extension, side bending, flexion, and rotation [7].Any issue associated with the lumbar spine can cause painful disorders including spinal stenosis, herniated disk, cervical spondylitis, kyphosis, and many more [45].An automated approach using machine learning (ML) and neural networks (NN) has been researched [9], [10], [11], [12], [13].Figure 2 highlights some of the commonly occurring lumbar spine disorders, including Hashimoto's disease, Graves' disease, and papillary carcinoma.
These are only some of the diseases that are caused by the lumbar spine, whereas its malfunctioning may cause some more serious and life-threatening diseases.Therefore, with the help of clinical tests and by exploring the application of technology, the implications and adverse effects of the diseases associated with it must be minimized.
The novelty of the research work in terms of feature extraction using the spider monkey approach and feature selection using linear differential optimization has been explored.The features extracted from the MRI images will be correlated with the categorical variable ''output.''The highest-correlated attributes with a minimum prescribed tolerance of 0.001 will be retained as a global solution.Upon termination of the iterations, the global solution with the minimum prescribed tolerance is retained.In the feature selection phase, the linearity model is used for fitting the extracted feature vector over the straight line.The feature values satisfying the linear equations are retained.Thus, the two-fold mechanism of feature extraction and selection allows the best possible features to be extracted and used for problem detection within the lumbar spine.This process helps increase the execution speed and classification accuracy of the operation.
The layout of the manuscript is structured as: Section II gives a systematic literature review with bibliometric analysis.Section III gives the problem identification and research motivation, describing the issues associated with existing literature and our proposal to resolve them.Section IV gives the proposed methodology along with the flow of the proposed work and algorithms.Section V gives the performance analysis and presents the obtained results.Section VI describes the contribution of the research.Section VII elaborates on the discussion.Section VIII concludes the work and provides its future scope.

II. LITERATURE REVIEW
This section elaborates on the bibliometric analysis and literature review of the selected papers.The bibliometric analysis is conducted to validate and select the research papers from reputed journals, including Web of Science, IEEE, Springer, and many more.Afterward, a literature review is conducted to find research gaps and to determine the best possible technique for LSD detection.The bibliometric analysis is conducted to select the papers for research on LSD detection.The process of bibliometric analysis for LSD detection is shown in Figure 3.
[14] discussed radiofrequency-based fusion surgery for handling the issues with the LSD.A dataset of MRI images was collected using real-time resources.[15] suggested that the disease was common among middle-aged people.This was validated using clinical testing procedures.This fact can 141890 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.be utilized for collecting a dataset for building an automated model to detect the disease.[16] discussed and presented a critical evaluation of classification models for the prediction of the disease.
ML-based mechanisms, including k-nearest neighbors (KNN), support vector machine (SVM), and RF, were used for comparison purposes.SVM achieved the highest classification accuracy of 90%.The authors [17] discussed the presence of transcriptome signatures in people with lumbar spine diseases.The level of signatures determined the severity of the disease [54].The clinical testing procedures were applied to prove the validity of the approach.The authors [18] discussed spinal problems in aging athletes.LSDs are common among athletes due to extra stretchability and extensibility.Some preventive measures must be in place to tackle the issue of the lumbar spine in athletics.
The research [19] explored the application of CNN for an early prediction of LSD using an MRI-based dataset derived from Kaggle.The authors [20] proposed the detection of lumbar hernias using fuzzy logic.The authors [21] discussed the application of soft computing and finite element methods for the detection of the disease.The execution speed was poor in this case.The authors [22] discussed the applications of VGG 16, ResNet50, and SE ResNet50 network models to detect the affected regions in MRI images.The authors [23] discussed an application of NNs for the detection of vertebral column disorders and lumbar disc disorders.The detection process with an NN-based approach applied to an MRI dataset presents a classification accuracy of 92%.The authors [24] proposed ML and deep learning (DL)-based models for the prediction of the disease and concluded that the two models could be employed for smaller and larger datasets respectively [52].Table 1 summarizes some important datasets and techniques used for LSD detection.
The dataset corresponding to the lumbar spine disorder was derived from Kaggle and used to analyze the lumbar spine disorder in the children's data set.A KNN-based classification was performed to achieve high classification accuracy within the range of 90%.[29] proposed a DL-based mechanism for detecting lumbar spine-related disease in children.Furthermore, individual disabilities were also detected using the CNN-based mechanism.The CNN produces high classification accuracy along with an f-score.The authors [30] proposed a recurrent neural network (RNN) [27] for lumbar spine disorder detection.An RNN-based network will allow the layers to establish connections with cycles or similar features that otherwise are not allowed within a CNN [53].The RNN-based model will help increase classification accuracy with the complex dataset.The dataset used in the proposed work consists of 500 MRI images as described in Table 2. Now the appraisal for internal and external validity is tested.Internal validity pertains to the robust methodology and comprehensive literature review.It is predominantly influenced by the degree of bias.The Newcastle-Ottawa Scale (NOS) is utilized for conducting internal validation tests [32].It delineates various measures related to selection bias, exposure, and comparability.These measures are employed for assessing the internal validity outcomes.The selection bias is minimal in this instance since the samples chosen for the examination in most of the literature were oriented toward people admitted to hospitals or through some benchmark websites.They might probably have the disease.Although certain individuals may not have the disease, selection bias is still negligible.A comprehensive literature review was carried out for the study.The study outlines allocation concealment before the patient group assignments, resulting in minimal selection bias.
The performance bias suggests uniform care across all patient groups.In this study, the patients suffering from the disease received preferential treatment compared to other patients.Hence, the study exhibits performance bias.This bias is certain to be present in the collected samples.The cause of this is that patients in the hospital may have different medical conditions.
Moreover, there is no evidence of detection bias in the study.The rationale for this is that the researcher is uninformed about the results of the interview and questionnaire.Determining the presence of detection bias is essential in evaluating subjective disease outcomes.The subjective observations show that there is an inverse connection between mental disorders and the chances of LSD.The study does not exhibit attrition bias, which relates to the absence of study data.During this analysis, we included and examined 359 cases and 708 controls without omitting any sample data.The presence of missing data can ultimately result in reduced classification accuracy.Since all the samples were thoroughly examined and missing values were addressed through a method based on the mean or mode, attrition bias is not a concern in this study.Reporting bias occurs when there's a notable contrast between the reported and unreported findings.However, in this study, reporting bias is non-existent, as there is no discernible distinction between the reported and unreported results.The outcomes obtained from this research align with prior studies, reinforcing the absence of reporting bias.
The evaluation of internal validity has been established through the NOS.Moreover, inaccuracies in data collection will detrimentally influence the outcome.Consequently, any errors could lead to complications in the classification of the samples, which consist of 359 cases and 708 controls.
Conversely, external validity suggests the potential applicability of the research findings to different individuals or settings.This aspect of validity encompasses the assessment of sample size, methodology, the studied population, and the research setting.The research incorporates a sample size of 359 cases and 708 controls.Every sample was obtained from a medical facility in Korea.To establish the authenticity of the collected sample, the nonprobability sampling technique was employed.The research employs convenience sampling, a nonprobability-based sampling approach chosen due to the convenient accessibility of the samples to the researcher [32].To assess the influence of vegetable and fruit intake on the likelihood of LSD and related diseases, accurate patient-related information is compulsory.To achieve this objective, both qualitative and quantitative research approaches can be utilized [33].Both interviews and questionnaires are used in the proposed research to 141892 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.collect population-specific data.To include the patients who were uncomfortable with in-person interviews, questionnaires were given to hospitalized patients at various geographic locations.Consequently, the study's target population comprised hospital-admitted patients, a highly suitable choice for this research.The research context revolves around identifying various preventive measures against LSDs and related conditions.To this end, this research investigated the influence of observing facial expressions and eye gaze on the likelihood of decreasing LSD and related conditions.It was discovered that LSD and related diseases were correlated with some of the facial expressions and eye gaze, confirming that there was a causal relationship between the exposure and the result.
Because LSD and other related disorders affect sizable populations all over the world, this study is accurate and effective for a large group of people.While this study pertains to children, its findings can be extended and applied globally.The exploratory investigation also revealed the fact that AI-based mechanisms can reduce the chances of LSD and other related diseases.The external validity of the research is proven because its overall result can be generalized to different populations throughout the world.

III. PROBLEM IDENTIFICATION AND RESEARCH MOTIVATION
The LSD detection-related issues discovered in the existing literature are in terms of feature extraction and selection [34].Extracted features using a NN-based approach do not show any correlation with the categorical outcome variable.
Consequently, feature selection is without correlation analysis.Overall, the degree of misclassification within such approaches was minimal.The pre-processing mechanism also does not include validation and normalization.Thus, the calculation complexity is high [35], [36], [37], [38], [39].Therefore, there is a strong motivation to perform this research.
Our research focuses on addressing the limitations found in existing research on LSD detection.While prior studies have explored various aspects of LSD detection, they often lack effective feature extraction and selection methods that capture the correlation between extracted features and the categorical output variable.Additionally, our work extends beyond pre-processing by incorporating validation and normalization steps, which are often overlooked in existing methodologies.By introducing a linearity-based model within the CNN framework for feature selection, we aim to enhance classification accuracy and reduce misclassification rates.These enhancements were not adequately addressed in prior research, and our work strives to bridge this gap.

IV. PROPOSED METHODOLOGY
The proposed work begins with the data acquisition phase; data is collected from both benchmarked and real-time datasets.In the pre-processing mechanism, any noise from the dataset is removed.Feature extraction and selection are applied to determine critical features for the prediction of LSD.Finally, classification is performed.The flow of the proposed system is shown in Figure 5.

A. PRE-PROCESSING
This phase is critical for the overall classification of the LSD.The pre-processing mechanism includes many sub-phases.The first phase is histogram equalization which enhances the contrast and overall clarity of the MRI images.After the enhancement process, median filtering is applied to remove salt and pepper noise.The validation mechanism is applied to test the restored image against the original image.If the restored image correlates with the original image, the restored image is retained.The pre-processing algorithm is given below as Algorithm 1.
The flowchart for preprocessing is given in Figure 6.

B. SEGMENTATION
Segmentation is the process of retaining only a critical section of the image and removing the unnecessary parts which is accomplished using background subtraction and region cutting.Background subtraction eliminates the background of the image and only the ROI will be retained.The ROI is carefully chosen to focus on the specific anatomical regions of interest within the lumbar spine.In this study, the ROI corresponds to the intervertebral disc (IVD) region, which is a critical component of the lumbar spine and can provide valuable insights into potential abnormalities or diseases.The IVD region is chosen as the ROI due to its significance for lumbar spine health and its susceptibility to various spinal For J = 1 to cols(Image) 6.
End of If 10.
End of For 11. End of For 12. Normalized = norm(Validated) conditions.IVDs act as shock absorbers between adjacent vertebrae and are composed of an inner nucleus pulposus surrounded by an outer annulus fibrosus.Any degeneration, herniation, or abnormality in the IVD can lead to severe back pain and related complications.
The selection of the IVD as the ROI is based on both anatomical and pathological considerations.It is the region where most spinal issues, such as herniated discs or degenerative changes, are likely to manifest.Therefore, analyzing this specific region can yield valuable diagnostic information.The ROI selection is further refined by considering radiological expertise, where a team of experienced radiologists or medical experts is involved in defining the precise boundaries of the IVD region to ensure accurate analysis.
The chosen ROI's boundaries are applied consistently across the dataset to maintain uniformity in analysis.This rigorous ROI selection process helps to focus the analysis on the most relevant and informative parts of the lumbar spine MRI images.It allows for effective feature extraction and subsequent classification, enhancing the accuracy of disease detection and contributing to the overall robustness of the study's methodology.This will be accomplished by reducing 255 (the maximum intensity of the pixels) from the intensity levels in the image: where ''I'' is the intensity level in the image.Back sub indicates the background-subtracted image.The region cut is Append(boundary_x, x) 9.
End of If 11.End of For 12. End of For applied to the background-subtracted image to extract the meaningful portion of the image using Algorithm 2.

C. FEATURE EXTRACTION
Feature extraction is applied over the extracted region.The process becomes faster as the entire image is not involved 141894 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
in feature extraction or feature vector generation.Feature extraction is accomplished using the SMO along with the digital differential analyzer approach.The utilization of SMO is advantageous in this scenario, illustrating two crucial facets of swarm intelligence: organization and task distribution.It takes into account both local and global solutions.In case of local solution proves to be superior to the global one, it will replace the global one.During the local leader phase, the local solution is derived, where the present search space is adjusted according to the progress of the phase.The local leader phase creates a path position to be explored by the SMO, signifying the feature's location in the MRI image as shown in Equation 1.To update the trial position in the sample space, the leadership position, the current energy position, and a randomly chosen entity within the group are used.The value of the perturbation rate determines the size of the solution space [12]., j)) where r(0, 1) ≥ Per_rate@SMO_pos(i1, j) where r(0, 1) where, 'r(0,1)' represents a random sample taken from the sample space, which also encompasses the perturbation rate.The use of a split and shift mechanism, in conjunction with perturbation, enables precise feature extraction.The objective function is influenced by factors such as throughput, energy efficiency, and the degree of imbalance.If the new solution surpasses the fitness values of the previous solution, it supersedes the existing solution.The fitness value associated with the present solution guides the process of node selection.The fitness function is represented in Equation (2).
The fitness equation above suggests that following the transmission of frames, each node must possess certain attributes; otherwise, the present solution is not valid.Equation (3) provides the objective function that supports the overall solution.
Here 'E c ' indicates the residual features that must be maximized, 'T' indicates throughput (number of features), and 'Im deg ' is the degree of imbalance that must be minimized.
The global leadership phase is critical and is responsible for updating the search space just like the local leadership phase but in a different manner as compared to the local leadership phase.The spider monkey (MRI image) gets updated or selected based on the probability value calculated by (4).
Probability Nselect = 0.9 * fitness n(i) Maxp−fitness cuurent + 0.1 (4) Maxp represents the peak expected fitness value and fitness current denotes the present fitness function value.// Verify the global solution (g g ) using f(n) 3.
While(fitness does not show enhancement) or (g < iteration max ) or (halting criteria) // Divide the population into 'k' groups 5.
// Conduct learning with consideration of features, throughput, and distance to assess fitness 7.
g g = g g +1 // Update the old solution to the new ones 10.
If (g g < g g +1) then // Select one solution (g) for selection 17.
// Allocate a position in the ranking for the viable solution node (t) = n e (g) 19.
End If 21.End While 22.Return node 23.End While Fitness n(i) represents the fitness score of the present node.The overall probability equals '1,' further divided into '0.9' and '0.1' to determine the likelihood of node selection.
After this, a new trial is repeated once more through the fitness equation.Then, the fitness score of the existing sensor node is compared with that of the newly computed node.The optimal sensor node is chosen for transmission using Algorithm 3. fit extracted value on straight line y = mx+b 5.
End of For 6. End of For

D. FEATURE SELECTION
The feature selection is based on a CNN along with a linearity model.The linearity model is included within the CNN.The extracted features will be fitted over a straight line regression line) using Algorithm 4. If the fitting is satisfied, features will be retained.Thus, the feature reduction is performed.
After performing all the steps of the methodology, the classifier is invoked [41].This model aids in selecting features that satisfy linear equations, enhancing the discriminative power of the selected features.

E. CLASSIFICATION
The classification is performed using an ensemble-based approach.The ensemble includes the MSVM, RF, DT, and NB approach.It was observed that MSVM has the lowest degree of misclassification and, hence, the highest classification accuracy.This result was obtained in correlation with the trained model.The trained model also performs well while testing with RF, DT, and NB approaches.

V. PERFORMANCE ANALYSIS AND RESULTS
The performance analysis corresponding to the different sections of methodology is presented in this section, including the result of pre-processing using histogram equalization, median filtering, and validation as shown in Figure 7.
We analyze the distribution of correlation values and determine a threshold based on the natural clustering or gaps in the data.We looked for inflection points in the distribution or considered percentiles for deciding threshold values.
The pre-processing improves the accuracy of the MRI image.The result of histogram equalization improves the contrast of the input image.The median filter is applied, which improves the overall clarity of the image by removing salt and pepper noise.The validity of the image after pre-processing is checked against the original image using correlation analysis.The outcome of the correlation analysis is given in Table 4.
From the validation analysis, it is observed that after preprocessing, features within the image are retained.
After pre-processing, background subtraction and region cut mechanisms are applied for segmentation.The segmentation-based mechanism removes the unnecessary parts from the image and retains only the critical parts.The result of the background subtraction and region cut is given in Figure 8.
After the segmentation, features are extracted using a CNN-based linearity model.The characteristics or features derived from the lumbar images are given in Table 5.
The plot for Table 5 is given in Figure 9.The visualization results indicate that the color features from the MRI images are the highest.
The result of classification using an MSVM, DT, NB, and RF approach is evaluated next.The evaluation in terms of classification accuracy, specificity, sensitivity, F-Score, and running time complexity is presented in Table 6.
141896 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.The high classification accuracy of 96% with MSVM is achieved with the optimization-based approach in the detection of LSD.Furthermore, high classification accuracy with other classifiers, including RF, NB, and DTs, also proves worth studying.The plot for Table 6 is given in Figure 10.The visualization results indicate that the MSVM validation metrics are the highest.
This section demonstrates that the results acquired are significantly more precise and transparent.The performance of the proposed technique is compared with RF, NB, and DT.The obtained results are given in Table 6 which demonstrate that 96 % classification accuracy is obtained with the proposed method.The second-best performance is obtained by RF (94%).DT and NB yielded accuracy values of 93.5% and 91%, respectively.Thus, the proposed technique outperforms all the compared methods.
The calculated sensitivity and specificity values are given in Table 6.As seen, 82% sensitivity and 85% specificity values were calculated for the proposed technique.These values are greater than those achieved with the other existing methods.The F-score for MSVM is 0.86.On the other hand, the F-score for RF, DT, and NB is 0.69, 0.59, and 0.68 respectively.

VI. CONTRIBUTION
The contribution section of the research paper outlines the novel aspects and advancements introduced by the authors in their study.It highlights the unique elements that the research brings to the field and explains how the proposed approach addresses existing limitations or challenges.Here's a summarized version of the contribution section based on the content you provided:

A. ENHANCED FEATURE EXTRACTION AND SELECTION
The proposed research introduces an optimized feature extraction and selection process that overcomes limitations in existing methods.Unlike traditional approaches, the feature extraction process is enhanced using an SMO technique, resulting in improved accuracy and correlation with the outcome variable.Additionally, feature selection employs a linearity-based model, further refining the selected features.

B. INTEGRATED PRE-PROCESSING MECHANISM
The research introduces a comprehensive pre-processing mechanism that includes histogram equalization, median filtering, validation, and normalization.This integrated approach effectively reduces noise and enhances the quality of MRI images, which positively impacts subsequent phases of the analysis.

C. SEGMENTATION FOR ROI DETECTION
The study incorporates a segmentation step that accurately identifies the ROI within MRI images.By utilizing background subtraction and region cut mechanisms, irrelevant portions of the image are eliminated, allowing focused analysis on critical areas related to lumbar spine issues.

D. ENSEMBLE-BASED CLASSIFICATION
The research employs an ensemble-based classification approach using SVM, RF, DTs, and NB algorithms.The classification outcomes from these techniques are compared, and the SVM model demonstrates the highest accuracy, further validating the effectiveness of the proposed approach.

E. VALIDATION AND GENERALIZATION:
The research meticulously addresses both internal and external validity through a structured approach.The utilization of the NOS for internal validity assessment ensures a rigorous evaluation of bias, while the consideration of different populations and contexts enhances external validity and generalizability.

F. INNOVATIVE APPLICATION OF SWARM INTELLIGENCE:
The integration of SMO [46], [48], a type of swarm intelligence, adds innovation to the feature extraction process.By simulating swarm behavior, this technique improves the accuracy and efficiency of feature extraction, contributing to overall enhanced disease detection.

G. COMPREHENSIVE LITERATURE REVIEW:
The paper includes an extensive literature review that not only supports the proposed approach but also identifies gaps and limitations in existing methods.This comprehensive analysis lays the foundation for the novelty and significance of the research.

H. PRACTICAL IMPLICATIONS:
The research holds practical implications for medical diagnosis and treatment.The accurate detection of lumbar spine issues through the proposed approach can lead to more timely and precise interventions, ultimately improving patient outcomes and quality of life.
In summary, the research makes significant contributions to the field of LSD detection through its innovative approach to feature extraction, comprehensive pre-processing, segmentation, classification, and validation.These contributions collectively enhance the accuracy, efficiency, and practical applicability of lumbar spine issue detection, addressing key challenges in the existing literature.

VII. DISCUSSION
In the discussion, the findings of this research underscore the innovative strides made in enhancing LSD detection.The optimized feature extraction and selection methods, particularly through the utilization of SMO and a linearity-based model, have exhibited promising results in terms of improving accuracy and correlation with outcomes.The integrated pre-processing approach, involving histogram equalization, median filtering, validation, and normalization, has effectively mitigated noise, enhancing the quality of MRI images [50].The inclusion of segmentation techniques, such as background subtraction and region cut, has further refined the analysis by isolating pertinent regions of interest.Ensemblebased classification, encompassing SVM, RF, DT, and NB, provides a comprehensive perspective on disease prediction.The validation efforts utilizing the NOS have fortified the internal and external validity of the study.Importantly, the application of swarm intelligence in the feature extraction process signifies a novel approach with promising implications for optimizing feature identification.Overall, our research contributes to the scientific understanding of LSD detection and also holds practical implications for medical diagnosis and treatment [51], potentially leading to better healthcare outcomes.Nevertheless, it is essential to acknowledge the evolving landscape of ML and medical imaging, including the emergence of transformer models [52], [53], [55], and suggest potential avenues for future research in refining and expanding the proposed approach to maintain its relevance and effectiveness in this dynamic field.

VIII. CONCLUSION AND FUTURE SCOPE
The lumbar spine can impact a person's life adversely.The painful nature of the spine can make the sufferer's life harder.To overcome the issue, a technology-driven approach is required.To this end, the proposed work uses an SMO-based differential analyzer for feature extraction.Feature selection is based on the CNN-based linearity model.Overall, both approaches allow for better fit and less loss in terms of classification.The ensemble-based approach is used to test the optimized model for the lumbar spine.High classification accuracy with MSVM, RF, DT, and NB proves the worth of the study.The test data for the proposed work is benchmarked and derived from online sources.In the future, a real-time dataset can be used along with the proposed optimization model to prove the validity of the approach.

FUNDING
This research work received no funding.
141898 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

FIGURE 1 .
FIGURE 1.The vital functions of the lumbar spine.
Selection A) Is the definition of the case satisfactory?a) Positive, backed by independent validation * b) Positive, such as via record linkage or self-reported data.c) Description lacking B) Are the cases well-represented a) continuous or typical sequence of instances * b) likelihood of selection biases or unspecified C) Selection of Controls a) community controls * b) hospital controls * c) absence of details D) Specification of Controls a) absence of prior disease record (endpoint) * b) absence of source details Comparability A) Equivalence of cases and controls determined by the study's design or analysis a) study controls for * LSD, Autism spectrum disorder (ASD), and attention deficit hyperactivity disorder (ADHD) b) study manages for any extra variable These variables, often referred to as individual characteristics or personal traits, have been described in Table 3. Factors: These can be considered as influencing factors or variables.Demographics: This term may encompass age, marriage status, education level, and sometimes family history.Lifestyle: This can encompass regular alcohol drinking and smoking habits as aspects of one's lifestyle.Socioeconomic Profile: This term may encompass education level and marriage status.Exposure A) Identification of exposure a) protected document (e.g., medical documentation) * b) standardized interview conducted with no knowledge of case/control status * c) non-blinded interview regarding case/control status d) self-written statement or medical documentation exclusively e) no description B) Uniform approach to identifying cases and controls a) yes * b) no C) Unresponsive rate a) identical rate for both groups * b) those who did not respond explained c) fluctuating rate with no designation

FIGURE 5 .
FIGURE 5. Flow of proposed work.

Algorithm 1
Pre-Processing (Image) Input: Image, the image for enhancement and noise removal.Output: Hist_Image, the image after histogram equalization.Filt_Image, the image after median filtering.corr is used to calculate the correlation.norm is used to find the normalized image.Validated indicates the validated image.Normalized indicates the normalized image.rows return the number of rows in the image.cols returns the number of columns in the image.Assumptions: Indices, I and J are the index variables.1. Input Image 2. Hist_Image = hist(image) 3. Filt_Image = median_filter(Hist_Image) 4. For I =1 to rows(Image) 5.

FIGURE 9 .
FIGURE 9. No. of from MRI images.

TABLE 1 .
Dataset and techniques applied for LSD detection.

TABLE 5 .
Features extracted from the linearity model.