In the current study, HN-MRI was found to be approximately 87% accurate in correctly diagnosing a T4a laryngeal tumor, which was consistent with the results of post laryngectomy histopathological examination. However, the study also indicated that there was variability in the accuracy of HN-MRI in identifying the extent of the tumor into adjacent anatomical subsites, regardless of whether the laryngeal tumor was primary or recurrent. Across the entire group of patients, HN-MRI was most accurate in identifying tumor invasion into the arytenoid and base of the tongue (more than 90% accuracy), while the lowest accuracy was in detecting tumor extension into the paraglottic space (45%). Additionally, HN-MRI was observed to be more specific than sensitive in detecting tumor extension into laryngeal subsites.

Accurate identification of the anatomic subsites involved in the larynx is crucial, and the use of the most recent staging edition of AJCC/TNM nomenclature impacts treatment selection and prognosis [2, 13,14,15]. Clinical evaluation is superior to HN-MRI in assessing the extent of early stage laryngeal tumors on the mucosal surface [14], whereas HN-MRI excels in assessing the deeper submucosal extent of the tumor, including involvement of the base of the tongue [6]. For instance, the involvement of the paraglottic and pre-epiglottic spaces may necessitate a change in treatment from definitive radiotherapy to concurrent chemoradiation (CRT), while cartilage penetration may require a change in treatment approach from functional preservation to total laryngectomy [2, 16]. In this study, there was a significant difference between the HN-MRI cT-category and the pT-category findings (p < 0.05) which led to over- or under-staging of cT-category (i.e. 12% of patients underwent unnecessary total laryngectomy due to over-staging of HN-MRI), which confirm the observation found by Kim and Nix [17, 18]. In the current study, the accuracy of HN-MRI in detecting supraglottic laryngeal cancer was 85%, which was higher than the accuracy found by Kim et al. (75%) [17]. This discrepancy may be due to the radiologists' cautious reporting approach that focused on the critical decision-making tumor extension to the laryngeal subsites and false interpretation caused by peritumoral inflammation [19]. To enhance the accuracy of HN-MRI, it's crucial to establish a new set of guidelines "laryngeal-RADs," which may standardize the interpretation of MRI results for laryngeal cancer and provide a clear implication of decision-making directed treatment approach [20].

Regarding paraglottic space infiltration, our study showed low sensitivity (20%) and high specificity (88%), compared with other studies that reported higher sensitivity (93–95%), but lower specificity (50–76%) [21, 22]. Detecting pre-epiglottic space infiltration is also difficult, as our study showed a sensitivity of only 32% but a high specificity of 95%. Other studies have reported higher sensitivity (100%) but lower specificity (85%) in this regard [21]. The challenge lies in accurately determining the boundary of primary tumor from the surrounding peri-tumoral inflammation, edema, and fibrosis [22]. The paraglottic space is best assessed on axial and coronal images and HN-MRI is superior to CT for assessment of paraglottic space [21, 22]. The pre-epiglottic space is best visible on axial and sagittal images [22]. These spaces are mostly composed of fat, which is bright on T1 and T2 sequences and shows no or only minimal contrast enhancement [22]. In contrast, laryngeal cancer show intermediate signal intensity on T1, T2, and STIR and moderate post contrast enhancement [22]. Compared to the tumor itself, peritumoral inflammation tends to show higher T2 signal, more intense contrast enhancement and less diffusion restriction [6, 7]. Relying more on DWI in such cases could have improved the concordance rates.

Regarding thyroid cartilage invasion, a study conducted by Taha et al., had reported higher sensitivity, specificity, positive predictive value, and negative predictive value of HN-MRI in detecting inner thyroid lamina invasion (93%, 82%, 88%, and 90%) [23] compared to our study's findings (59%, 75%, 77%, and 55%, respectively). Also, Taha et al., had reported a higher sensitivity, specificity, positive predictive value, and negative predictive value of 85% for full thickness thyroid cartilage invasion [23] compared to our study's findings (74%, 61%, 79%, and 79%, respectively). Li et al., had reported that neck CT scans had a lower sensitivity, specificity, positive predictive value, and negative predictive value (57%, 86%, 65%, and 81%, respectively) compared with HN-MRI ( 94%, 87%, 78%, and 97%, respectively) [24]. Our study showed lower HN-MRI sensitivity, specificity, positive predictive value, and negative predictive value of (74%, 61%, 76%, and 58%, respectively) compared to Li et al., study. Perhaps, relying on multiparametric MRI assessment could improve the agreement rates in our study, which can differentiate between normal and ossified cartilage, peritumoral inflammation, and tumoral invasion [6, 7]. Neck CT scan cannot reliably make that distinction between the normal and ossified cartilage, sclerosis due to tumor invasion and sclerosis due to peritumoral inflammation [7].

Differentiating between cT3 and cT4 holds critical significance for the management and prognosis of laryngeal cancer patients [2, 25]. Tumors labeled as cT4a due to cartilage penetration exhibit heightened local recurrence risk post functional preservation. The VA-larynx randomized clinical trial revealed that salvage laryngectomy was necessary for 44% of Stage IV laryngeal cancer patients compared to 29% of Stage III (P = 0.048) and 56% of T4 tumors (those with cartilage penetration) vs 29% of smaller primary tumors (P = 0.001) [25]. Notably, RTOG 91–11 excluded cT4 patients due to the increased local failure risk observed in the VA-RCT [2, 25]. Within our clinical practice guidelines, we recommend total laryngectomy for patients with T4 disease due to cartilage penetration, substantial bulky T3 tumors with a non-functioning larynx, and as a salvage option following recurrence or persistent disease post functional preservation treatment [15]. Our study showed that 10/78 (12.8%) patients categorized as cT4, and without prior laryngeal treatment, underwent unnecessary primary laryngectomy and 1/4 (25%) categorized as cT3 disease found to have cT2 disease after laryngectomy who was candidate for functional preservation approach.

Limitations of this study include its retrospective nature. The study has inherent bias because we only reviewed cases that went to laryngectomy and missing true overall population sensitivity/specificity/accuracy in all comers, furthermore, the study included only those who had disease that was determined to be surgical. Incomplete radiology reports can also limit the study's findings as not all laryngeal subsites were reported. This means that the study may miss some important details that could affect the interpretation of the results. In addition, variability of MR technique and different MR scanners used for imaging the patients may also be contributing the study limitation. Furthermore, the lack of use of DWI/ADC in a substantial proportion of patients. However, our study reports on a relatively large cohort of a homogenous group of patients who received the same treatment protocol. Imaging the larynx with MRI is challenging mainly because of the long acquisition time and motion artifact induced by breathing, swallowing, and vessel pulsation but this has improved with newer techniques, including parallel imaging, and equipment [26]. The advancement in technology have also led to significant improvements in the image quality and spatial resolution, which can lead to more reliable assessment of cartilage invasion for example [7]. The addition of DWI to the magnetic resonance protocol has the potential to increase the specificity of the diagnosis for cartilage involvement and for discrimination of peritumoral edema from neoplastic tissue when assessing paraglottic space invasion [23, 27]. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) can help better differentiate between recurrent malignant tumors after CRT and changes caused by CRT [28]. DCE-MRI metrics can also predict response to CRT and potentially guide treatment [29,30,31]. Combined, parallel imaging and other technique have significantly improved the imaging acquisition times so that high temporal resolution cine MR imaging of the larynx is now possible at 3 T, which can demonstrate and quantify the intrafractional HN tumor motion and can help improve radiation planning and delivery [32, 33].