[Meningiomas of the cavernous sinus: the surgical resectability and complications]. / Meningiomas do seio cavernoso: correlação entre a extensão de ressecção cirúrgica e lesões neurovasculares em 16 pacientes.

The optimal management for patients with cavernous sinus meningiomas is to evacuate tumor without causing mortality or morbidity. The records of 16 patients, including 11 women and 5 men ranging in age from 31 to 63 years, underwent surgical treatment for this condition were reviewed. Completeness of tumor resection, cranial nerve morbidity, complications, mortality, the internal carotid artery encasement and outcome were studied. Total removal was achieved in six patients. Of ten patients who underwent subtotal resection there was one death and four were sent to radiotherapy. Morbidity was 24% for cranial nerves controlling extraocular motor function; trigeminal nerve function did not improve after surgical treatment. Symptomatic recurrence occurred in two patients who underwent subtotal tumor resection and in one who underwent complete tumor resection. The average follow-up period was of 26 months. According to our findings, we conclude: 1) the resectability of meningiomas of cavernous sinus depends on the degree of internal carotid artery involvement; 2) total resection of meningiomas confined in cavernous sinus is rare; 3) morbidity of the cranial nerves is significant; 4) subtotal resection is an effective mean to obtain control of the disease.

Autonomous learning based on cost assumptions: theoretical studies and experiments in robot control.

Autonomous learning techniques are based on experience acquisition. In most realistic applications, experience is time-consuming: it implies sensor reading, actuator control and algorithmic update, constrained by the learning system dynamics. The information crudeness upon which classical learning algorithms operate make such problems too difficult and unrealistic. Nonetheless, additional information for facilitating the learning process ideally should be embedded in such a way that the structural, well-studied characteristics of these fundamental algorithms are maintained. We investigate in this article a more general formulation of the Q-learning method that allows for a spreading of information derived from single updates towards a neighbourhood of the instantly visited state and converges to optimality. We show how this new formulation can be used as a mechanism to safely embed prior knowledge about the structure of the state space, and demonstrate it in a modified implementation of a reinforcement learning algorithm in a real robot navigation task.

[Subfrontal approach in sellar and suprasellar lesions]. / Acesso sub-frontal a lesões selares e para-selares.

A variety of mass lesions are placed in and around of the sella turcica. Although their common location these lesions have distinguishing features and different approaches can be selected. There are conventional transcranial and sophisticated skull base approaches. We report our experience with a subfrontal route to the sellar region. Between February 1997 and March 1998, 19 tumors placed around the sella have been treated surgically by this method. Eleven of them were pituitary tumors, 3 were craniopharyngeomas, 2 were meningiomas, and 1 was germinoma. Total removal was achieved in 17 cases. There was not operative mortality. Postoperative complications included twelve cases of unilateral anosmia and four cases of transitory diabetes insipidus. Progressive stroke occurred in one patient. These results stress the importance of the subfrontal approach to reach such lesions in sellar region.

Autonomous learning based on cost assumptions: theoretical studies and experiments in robot control.

Autonomous learning techniques are based on experience acquisition. In most realistic applications, experience is time-consuming: it implies sensor reading, actuator control and algorithmic update, constrained by the learning system dynamics. The information crudeness upon which classical learning algorithms operate make such problems too difficult and unrealistic. Nonetheless, additional information for facilitating the learning process ideally should be embedded in such a way that the structural, well-studied characteristics of these fundamental algorithms are maintained. We investigate in this article a more general formulation of the Q-learning method that allows for a spreading of information derived from single updates towards a neighbourhood of the instantly visited state and converges to optimality. We show how this new formulation can be used as a mechanism to safely embed prior knowledge about the structure of the state space, and demonstrate it in a modified implementation of a reinforcement learning algorithm in a real robot navigation task.