From pixels to prognosis: unveiling radiomics models with SHAP and LIME for enhanced interpretability.

Radiomics-based prediction models have shown promise in predicting Radiation Pneumonitis (RP), a common adverse outcome of chest irradiation. Τhis study looks into more than just RP: it also investigates a bigger shift in the way radiomics-based models work. By integrating multi-modal radiomic data, which includes a wide range of variables collected from medical images including cutting-edge PET/CT imaging, we have developed predictive models that capture the intricate nature of illness progression. Radiomic features were extracted using PyRadiomics, encompassing intensity, texture, and shape measures. The high-dimensional dataset formed the basis for our predictive models, primarily Gradient Boosting Machines (GBM)-XGBoost, LightGBM, and CatBoost. Performance evaluation metrics, including Multi-Modal AUC-ROC, Sensitivity, Specificity, and F1-Score, underscore the superiority of the Deep Neural Network (DNN) model. The DNN achieved a remarkable Multi-Modal AUC-ROC of 0.90, indicating superior discriminatory power. Sensitivity and specificity values of 0.85 and 0.91, respectively, highlight its effectiveness in detecting positive occurrences while accurately identifying negatives. External validation datasets, comprising retrospective patient data and a heterogeneous patient population, validate the robustness and generalizability of our models. The focus of our study is the application of sophisticated model interpretability methods, namely SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations), to improve the clarity and understanding of predictions. These methods allow clinicians to visualize the effects of features and provide localized explanations for every prediction, enhancing the comprehensibility of the model. This strengthens trust and collaboration between computational technologies and medical competence. The integration of data-driven analytics and medical domain expertise represents a significant shift in the profession, advancing us from analyzing pixel-level information to gaining valuable prognostic insights.

The contribution of diffusion tensor imaging and magnetic resonance spectroscopy for the differentiation of breast lesions at 3T.

BACKGROUND: Conventional breast magnetic resonance imaging (MRI), including dynamic contrast-enhanced MR mammography (DCE-MRM), may lead to ambiguous diagnosis and unnecessary biopsies. PURPOSE: To investigate the contribution of proton MR spectroscopy (1H-MRS) combined with diffusion tensor imaging (DTI) metrics in the discrimination between benign and malignant breast lesions. MATERIAL AND METHODS: Fifty-one women with known breast abnormalities from conventional imaging were examined on a 3T MR scanner. DTI was performed during breast MRI, and fractional anisotropy (FA) and apparent diffusion coefficient (ADC) were measured in the breast lesions and the contralateral normal breast. FA and ADC were compared between malignant lesions, benign lesions, and normal tissue. 1H-MRS was performed after gadolinium administration and choline peak was qualitatively evaluated. RESULTS: In our study 1H-MRS showed a sensitivity of 93.5%, specificity 80%, and accuracy 88.2%. FA was significantly higher in breast carcinomas compared to benign lesions. However, no significant difference was observed in ADC between benign and malignant lesions. The combination of Cho presence and FA achieved higher levels of accuracy and specificity in discriminating malignant from benign lesions over Cho presence or FA alone. CONCLUSION: In conclusion, applying DTI and 1H-MRS together, adds incremental diagnostic value in the characterization of breast lesions and may sufficiently improve the low specificity of conventional breast MRI.

Application value of 3T ¹H-magnetic resonance spectroscopy in diagnosing breast tumors.

BACKGROUND: Assessment of breast lesions with magnetic resonance imaging (MRI) provides a means for lesion detection and diagnosis. Proton (hydrogen-1) magnetic resonance spectroscopy ((1)H-MRS) has been proposed as a useful diagnostic technique in providing metabolic information of suspicious breast lesions. PURPOSE: To determine the clinical significance of in-vivo single voxel (1)H-MRS at 3T in the assessment of benign and malignant breast lesions in combination with dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). MATERIAL AND METHODS: Twenty-four women with known breast abnormalities from conventional imaging (mammography, ultrasonography) underwent DCE-MRI at a 3T MR scanner and 26 breast lesions were detected. Breast lesions were assessed according BI-RADS classification. Single voxel (1)H-MRS was performed after gadolinium administration and choline peak was qualitatively evaluated. All lesions were confirmed histologically from the surgically excised specimens. Sensitivity, specificity, and accuracy of the (1)H-MRS, of the BI-RADS classification and of their combination (DCE-MRI + (1)H-MRS) were calculated. RESULTS: Fifteen out of 26 lesions proved to be malignant and 11 proved to be benign. In our study (1)H-MRS showed sensitivity 80%, specificity 81.8%, and accuracy 80.7%. DCE-MRI showed sensitivity 100%, specificity 63.6%, and accuracy 84.6%. The combination of DCE-MRI and (1)H-MRS provided higher accuracy (96.4%), as well as higher specificity 81.8% compared to BI-RADS classification. CONCLUSION: The combined use of (1)H-MRS and DCE-MRI found to have improved diagnostic performance in the assessment of equivocal breast lesions. (1)H-MRS can be used as a useful adjunct during lesion characterization in clinical routine in cases classified as BI-RADS 3 and 4.

Impact of beam quality on megavoltage radiotherapy treatment techniques utilizing gold nanoparticles for dose enhancement.

This study determines the optimal clinical scenarios for gold nanoparticle dose enhancement as a function of irradiation conditions and potential biological targets using megavoltage x-ray beams. Four hundred and eighty clinical beams were studied for different potential cellular or sub-cellular targets. Beam quality was determined based on a 6 MV linac with and without a flattening filter for various delivery conditions. Dose enhancement ratios DER = D(GNP)/D(water) were calculated for all cases using the GEANT4 Monte Carlo code and the CEPXS/ONEDANT radiation transport deterministic code. Dose enhancement using GEANT4 agreed with CEPXS/ONEDANT. DER for unflattened beams is ∼2 times larger than for flattened beams. The maximum DER values were calculated for split-IMRT fields (∼6) and for out-of-field areas of an unflattened linac (∼17). In-field DER values, at the surface of gold nanoparticles, ranged from 2.2 to 4.2 (flattened beam) and from 3 to 4.7 (unflattened beams). For a GNP cluster with thicknesses of 10 and 100 nm, the DER ranges from 14% to 287%. DER is the greatest for split-IMRT, larger depths, out-of-field areas and/or unflattened linac. Mapping of a GNP location in tumor and normal tissue is essential for efficient and safe delivery of nanoparticle-enhanced radiotherapy.

Distinct peak at 3.8 ppm observed by 3T MR spectroscopy in meningiomas, while nearly absent in high-grade gliomas and cerebral metastases.

The purpose of the present study was to evaluate distinct metabolic features of meningiomas to distinguish them from other brain lesions using proton magnetic resonance spectroscopy. The study was performed on 17 meningiomas, 24 high-grade gliomas and 9 metastases. Elevated signal intensity at 3.8 ppm observed in low TE spectra adequately differentiated meningioma from other brain tumors while alanine was not indicative of meningioma occurrence; the presence of lipids and lactate did not provide a strong index for meningioma malignancy.