Comparing supervised and semi-supervised machine learning approaches in NTCP modeling to predict complications in head and neck cancer patients.

Background and purpose: Head and neck cancer (HNC) patients treated with radiotherapy often suffer from radiation-induced toxicities. Normal Tissue Complication Probability (NTCP) modeling can be used to determine the probability to develop these toxicities based on patient, tumor, treatment and dose characteristics. Since the currently used NTCP models are developed using supervised methods that discard unlabeled patient data, we assessed whether the addition of unlabeled patient data by using semi-supervised modeling would gain predictive performance. Materials and methods: The semi-supervised method of self-training was compared to supervised regression methods with and without prior multiple imputation by chained equation (MICE). The models were developed for the most common toxicity outcomes in HNC patients, xerostomia (dry mouth) and dysphagia (difficulty swallowing), measured at six months after treatment, in a development cohort of 750 HNC patients. The models were externally validated in a validation cohort of 395 HNC patients. Model performance was assessed by discrimination and calibration. Results: MICE and self-training did not improve performance in terms of discrimination or calibration at external validation compared to current regression models. In addition, the relative performance of the different models did not change upon a decrease in the amount of (labeled) data available for model development. Models using ridge regression outperformed the logistic models for the dysphagia outcome. Conclusion: Since there was no apparent gain in the addition of unlabeled patient data by using the semi-supervised method of self-training or MICE, the supervised regression models would still be preferred in current NTCP modeling for HNC patients.

Mucositis reduction by selective elimination of oral flora in irradiated cancers of the head and neck: a placebo-controlled double-blind randomized study.

PURPOSE: The aim of the study was to test the hypothesis that aerobic Gram-negative bacteria (AGNB) play a crucial role in the pathogenesis of radiation-induced mucositis; consequently, selective elimination of these bacteria from the oral flora should result in a reduction of the mucositis. METHODS AND MATERIALS: Head-and-neck cancer patients, when scheduled for treatment by external beam radiation therapy (EBRT), were randomized for prophylactic treatment with an oral paste containing either a placebo or a combination of the antibiotics polymyxin E, tobramycin, and amphotericin B (PTA group). Weekly, the objective and subjective mucositis scores and microbiologic counts of the oral flora were noted. The primary study endpoint was the mucositis grade after 3 weeks of EBRT. RESULTS: Seventy-seven patients were evaluable. No statistically significant difference for the objective and subjective mucositis scores was observed between the two study arms (p = 0.33). The percentage of patients with positive cultures of AGNB was significantly reduced in the PTA group (p = 0.01). However, complete eradication of AGNB was not achieved. CONCLUSIONS: Selective elimination of AGNB of the oral flora did not result in a reduction of radiation-induced mucositis and therefore does not support the hypothesis that these bacteria play a crucial role in the pathogenesis of mucositis.

Beam intensity modulation using tissue compensators or dynamic multileaf collimation in three-dimensional conformal radiotherapy of primary cancers of the oropharynx and larynx, including the elective neck.

INTRODUCTION: The treatment of midline tumors in the head and neck by conventional radiotherapy almost invariably results in xerostomia. This study analyzes whether a simple three-dimensional conformal radiotherapy (3D-CRT) technique with beam intensity modulation (BIM) (using a 10-MV beam of the MM50 Racetrack Microtron) can spare parotid and submandibular glands without compromising the dose distribution in the planning target volume (PTV). METHODS: For 15 T2 tumors of the tonsillar fossa with extension into the soft palate (To) and 15 T3 tumors of the supraglottic larynx (SgL), conventional treatment plans, consisting of lateral parallel opposed beams, were used for irradiation of both the primary tumor (70 Gy) and the elective neck regions (46 Gy). Separately, for each tumor a 3-D conformal treatment plan was developed using the 3-D computer planning system, CadPlan, and Optimize, a noncommercial program to compute optimal beam profiles. Beam angles were selected with the intention of optimal sparing of the salivary glands. The intensity of the beams was then modulated to achieve a homogeneous dose distribution in the target for the given 3D-CRT techniques. The dose distributions, dose-volume histograms (DVHs) of target and salivary glands, tumor control probabilities (TCPs), salivary gland volumes absorbing a biologically equivalent dose of greater than 40 or 50 Gy, and normal tissue complication probabilities (NTCPs) of each treatment plan were computed. The parameters of the 3D-CRT plans were compared with those of the conventional plans. RESULTS: In comparison with the conventional technique, the dose homogeneity in the target volume was improved by the conformal technique for both tumor sites. In addition, for the SgL conformal technique, the average volumes of the parotid glands absorbing a BED of greater than 40 Gy (V40) decreased by 23%, and of the submandibular glands by 7% (V40) and 6% (V50). Consequently, the average NTCPs for the parotid and submandibular glands were reduced by 7% and 6%, respectively. For the To conformal techniques, the V40 of the parotid glands was decreased on average by 31%, resulting in an average reduction of the NTCP by 49%. Both the average V50 and the NTCP of the submandibular glands were decreased by 7%. CONCLUSION: For primary tumors of the oropharynx, the parotid glands could be spared to a considerable degree with the 3D-CRT technique. However, particularly the ipsilateral submandibular gland could not be spared. For primary tumors of the larynx, the 3D-CRT technique allows sparing of all salivary glands to a considerable and probably clinically relevant degree. Moreover, the conformal techniques resulted in an increased dose homogeneity in the PTV of both tumor sites.

A simplified CT-based definition of the lymph node levels in the node negative neck.

INTRODUCTION AND PURPOSE: Using three dimensional (3D) conformal radiotherapy (CRT) techniques for elective neck irradiation (ENI) may allow for local disease control to be maintained while diminishing xerostomia by eliminating major salivary glands (or parts thereof) from the treatment portals. The standardization of CT based target volumes for the clinically negative (elective) neck is a prerequisite for 3DCRT. The aim of the present study was to substantially modify an existing ('original') CT-based protocol for the delineation of the neck target volume, into a more practical ('simplified') protocol. This will allow for rapid contouring and the implementation of conformal ENI in routine clinical procedures. MATERIAL AND METHODS: An earlier ('original') version of the CT-based definition for elective neck node regions 2-5 was re-evaluated, using 15 planning CT scans of previously treated patients. The contouring guidelines were simplified by (1) using a smaller number of easily identifiable soft tissue- and bony anatomical landmarks, which in turn had to be identified in only a limited number of CT slices, and (2) by subsequently interpolating the contoured lymph node regions. The adequacy of target coverage and the sparing using both 'original' and 'simplified' delineation protocols was evaluated by DVH analysis after contouring the primary tumor, the neck and the major salivary glands in a patient with supraglottic laryngeal (SGL) carcinoma who was treated using a 3DCRT technique. RESULTS: The BEV projections of the 'original' and the 'simplified' versions of the 3D elective neck target showed good agreement and were found to be reproducible. The DVH's of the target and parotid glands were not significantly different using both contouring protocols. CONCLUSIONS: The 'simplified' protocol for the delineation of the 3D elective neck target produced both comparable target coverage and sparing of the major salivary glands. When used together with an interpolation program, this 'simplified' protocol substantial reduced the contouring time and makes ENI with sparing of the major salivary glands a practical and achievable goal.

A three-dimensional CT-based target definition for elective irradiation of the neck.

INTRODUCTION: Elective treatment of the clinically node-negative neck by radiation results in excellent control rates. However, radiation therapy with its organ-preserving properties is not without morbidity. Side effects of elective neck irradiation are mainly due to damage of the major and minor salivary glands, resulting in the dry mouth syndrome. Given that RT is the preferred treatment modality in case of elective treatment of the neck in many institutions, it is of utmost importance to try and reduce the associated sequelae of RT. MATERIAL AND METHODS: With the introduction of CT-planning systems and the development of 3D conformal radiation therapy (3D CRT) techniques, it has become feasible to deliver adequate doses of radiation to the target (neck) and at the same time saving (parts of) the salivary glands from doses beyond tolerance. A prerequisite for these techniques is that they require a precise knowledge of the target (i.e., of the elective neck) on CT. To be able to correlate borders of the surgical levels in the neck (I-VI) with structures seen on CT, an anatomical study, using two fixed (phenol, formaldehyde) human cadavers, was performed. Subsequently, the 6 potential lymph node regions in the neck on CT were defined. RESULTS AND DISCUSSION: The reference for the current 3D CT-based definition of the lymph node regions in the neck is the official report of the American Academy of Otolaryngology, describing, based on surgical anatomy, the lymph node groups in the neck by Levels I-VI. The present investigation depicts reproducible landmarks on transversal CT images, corresponding to anatomical reference structures known from surgical levels (I-VI) and, this way, CT-based lymph node regions (1-6) were constructed.

Radiation-induced bilateral optic neuropathy in cancer of the nasopharynx. Case failure analysis and a review of the literature.

CASE REPORT: A case history of unanticipated radiation-induced bilateral optic neuropathy, 18 months after induction chemotherapy and radiation therapy for a locally advanced nasopharyngeal carcinoma, is presented. Retrospective reanalysis of the radiation therapy technique, with emphasis on the doses received by the optic pathway structures, was performed. These re-calculations revealed unexpectedly high doses in the range 79 to 82 Gy (cumulative external and brachytherapy dose) at the level of the optic nerves, which explained the observed radiation injury. CONCLUSION: Routine implementation of computed tomography for 3D dose planning purposes is therefore advocated. Review of the current literature confirms the importance of 3D dose planning in avoiding this complication and high-lights the role of MRI in establishing the diagnosis of radiation-induced optic neuropathy.

Treatment of faecal incontinence in a child with sacral agenesis: the use of biofeedback training.

The case of a 14-year-old girl is presented who was born with sacral agenesis and severe, chronic constipation not responding to laxative treatment. Biofeedback was employed in the treatment of the constipation and soiling. After six treatment sessions over four months, the constipation and soiling ceased and there was no relapse during follow-up of 12 months. Although this child had an absence of S3-S5 and the coccyx, with possible innervation abnormalities, this case illustrates that biofeedback training might be a useful adjunct to existing therapies.

Staging and prognosis using endosonography in patients with inoperable esophageal carcinoma treated with combined intraluminal and external irradiation.

Brachytherapy and external irradiation combined are an alternative to surgery in the treatment of advanced esophageal cancer. Endosonography has proved to be an accurate method of staging the depth of tumor invasion of esophageal cancer. Sixty-three patients with inoperable esophageal cancer underwent endosonography followed by combined brachytherapy and external irradiation. Staging was incomplete in 31 of 63 patients because of tight stenosis or difficulty in imaging celiac lymph nodes. During follow-up of 23 patients, reduction of tumor thickness and lymph node abnormalities was observed in 16. The median survival was 10.4 months. Survival time was correlated with initial number of metastatic lymph nodes found by endosonography. Paradoxically, the survival of patients with more extensive intra-luminal tumor growth was significantly better than survival of those with less tumor growth. In summary, the response to combined brachytherapy and external irradiation could accurately be assessed with endosonography. Lymph node abnormalities and tumor thickness were closely related to survival rate.

Manometry, profilometry, and endosonography: normal physiology and anatomy of the anal canal in healthy children.

Normal and manometric and profilometric values and normal endosonographic features of the anal canal are required for evaluation of pathological conditions such as slow-transit constipation, anorectal outlet obstruction, and Hirschsprung's disease, status after surgery for imperforate anus, and other anal abnormalities. Anorectal manometry, profilometry (rapid-pull-through, three-dimensional, eight-channel radial manometry), and endosonography were carried out in 13 healthy children. A significant correlation was demonstrated between conventional manometric and profilometric maximal squeeze pressure of the external anal sphincter (EAS). However, the maximal and resting tone of the sphincter complex in profilometry was twice as high as in manometry due to reflexive contraction of a pelvic floor muscle, probably the EAS. With profilometry a positive correlation was found between age and sphincter length. Endosonographically assessed thickness of the EAS, puborectal muscle, and levator ani complex showed a significant correlation with age. However, no correlation was demonstrated between age and internal anal sphincter thickness. Thus, the development of the essential structures of the anal canal in children is age dependent. In addition, these measurements of normal physiology and anatomy of the anal canal provide the basis for detecting pathological conditions of the anorectal region in children.

Endosonography in the clinical staging of Klatskin tumor.

Endosonography was performed preoperatively in 46 patients with carcinoma of the common hepatic duct and its bifurcation. The results of endosonography were correlated with findings during surgery and pathological examination of the resected specimen and classified according to the new (1987) TNM classification. Overall accuracy in assessing the depth of tumor infiltration was 86.0%. Endosonography was accurate in predicting the presence of lymph nodes but not accurate in defining non-metastatic changes of lymph nodes. Staging of distant metastases was not accurate due to the low penetration depth of ultrasound.

Endosonographic TNM staging of extrahepatic bile duct cancer: comparison with pathological staging.

Endosonography was performed preoperatively in 33 patients with common bile duct carcinoma and in 43 patients with carcinoma of the common hepatic duct and its bifurcation. The results were correlated with the histology of resected specimens according to the new (1987) TNM (tumor, node, metastasis) classification. Endosonography was accurate in the evaluation of the depth of tumor infiltration. Overall accuracy for common bile duct carcinoma and common hepatic duct carcinoma was 82.8% and 85%, respectively. Endosonography was helpful in diagnosing regional lymph node metastases but not accurate in diagnosing nonmetastatic lymph nodes. With common bile duct carcinoma, the incidence of lymph node metastasis increased with progressive depth of tumor infiltration. No such correlation was found in common hepatic duct carcinomas. In the staging of distant metastasis, this technique was limited by the low-penetration depth of ultrasonography. Thus, additional transcutaneous ultrasonography or computed tomography was necessary for complete staging. The routine use of the biopsy channel for endosonographically guided aspiration puncture will further enhance the diagnostic value of endosonography in the future.

Endosonography of peri-anal and peri-colorectal fistula and/or abscess in Crohn's disease.

Transcolorectal endosonography (ES) was performed in 36 patients with Crohn's disease suspected clinically to have a fistula or abscess. A hypoechoic or anechoic duct-like lesion immediately adjacent to the anorectal lumen compatible with a fistula was found in 32 patients. A communication between the fistulous tract and adjacent structures such as the skin, anal canal, or vagina was detected in all 32 patients. An anechoic cavity adjacent to or communicating with the fistula was visualized in 29 of the 36 patients. A fistula was visualized in the remaining seven patients with no evidence of an abscess. This anechoic cavity compatible with an abscess was surgically confirmed in 14 of 17 patients. We judged the extent and configuration of the abnormalities to be more clearly visualized by ES when results were compared with fistulography in five patients. There were no ES complications, and we conclude that ES is the preferred diagnostic procedure in patients with peri-rectal pathology because of the low risk of bacterial dissemination and low incidence of patient discomfort. Utilizing ES after non-surgical treatment was successful in 19 patients for documentation of the response to therapy.

Preoperative TNM classification of proximal extrahepatic bile duct carcinoma by endosonography.

ES was performed preoperatively in 39 patients with a proximal bile duct carcinoma. The results were correlated with the histologic findings of resected specimens or intraoperative excision according to the new (1987) TNM classification. ES was accurate in assessing the depth of tumor infiltration. The overall accuracy of ES was 83.8%. Overstaging of T2 carcinoma occurred in three of ten patients (30%) and understaging of T3 carcinoma occurred in 3 of 27 patients (11%). ES was helpful in diagnosing lymph node metastasis (accuracy, 92.9%) but not accurate in defining nonmetastatic lymph nodes (accuracy, 18.8%). Difficulties occurred in distinguishing inflammatory lymph nodes and micrometastatic lymph node involvement. ES was not accurate in diagnosing hepatic metastases and peritoneal dissemination because of the limited penetration depth of ultrasound. The incidence of lymph node metastasis of advanced stage carcinoma was approximately 40%. Technical improvements, such as reduction of the diameter of the echoprobe, easy handling of ES-guided cytologic puncture, and the routine use of a catheter echoprobe during ERCP will further enhance the accuracy of ES.